staging: rtl8188eu: rename HalSetBrateCfg() - style
[linux/fpc-iii.git] / drivers / edac / sb_edac.c
blob07726fb00321e8cb5a9e53a9d73ac01cfbbc1cfa
1 /* Intel Sandy Bridge -EN/-EP/-EX Memory Controller kernel module
3 * This driver supports the memory controllers found on the Intel
4 * processor family Sandy Bridge.
6 * This file may be distributed under the terms of the
7 * GNU General Public License version 2 only.
9 * Copyright (c) 2011 by:
10 * Mauro Carvalho Chehab
13 #include <linux/module.h>
14 #include <linux/init.h>
15 #include <linux/pci.h>
16 #include <linux/pci_ids.h>
17 #include <linux/slab.h>
18 #include <linux/delay.h>
19 #include <linux/edac.h>
20 #include <linux/mmzone.h>
21 #include <linux/smp.h>
22 #include <linux/bitmap.h>
23 #include <linux/math64.h>
24 #include <linux/mod_devicetable.h>
25 #include <asm/cpu_device_id.h>
26 #include <asm/intel-family.h>
27 #include <asm/processor.h>
28 #include <asm/mce.h>
30 #include "edac_module.h"
32 /* Static vars */
33 static LIST_HEAD(sbridge_edac_list);
36 * Alter this version for the module when modifications are made
38 #define SBRIDGE_REVISION " Ver: 1.1.2 "
39 #define EDAC_MOD_STR "sb_edac"
42 * Debug macros
44 #define sbridge_printk(level, fmt, arg...) \
45 edac_printk(level, "sbridge", fmt, ##arg)
47 #define sbridge_mc_printk(mci, level, fmt, arg...) \
48 edac_mc_chipset_printk(mci, level, "sbridge", fmt, ##arg)
51 * Get a bit field at register value <v>, from bit <lo> to bit <hi>
53 #define GET_BITFIELD(v, lo, hi) \
54 (((v) & GENMASK_ULL(hi, lo)) >> (lo))
56 /* Devices 12 Function 6, Offsets 0x80 to 0xcc */
57 static const u32 sbridge_dram_rule[] = {
58 0x80, 0x88, 0x90, 0x98, 0xa0,
59 0xa8, 0xb0, 0xb8, 0xc0, 0xc8,
62 static const u32 ibridge_dram_rule[] = {
63 0x60, 0x68, 0x70, 0x78, 0x80,
64 0x88, 0x90, 0x98, 0xa0, 0xa8,
65 0xb0, 0xb8, 0xc0, 0xc8, 0xd0,
66 0xd8, 0xe0, 0xe8, 0xf0, 0xf8,
69 static const u32 knl_dram_rule[] = {
70 0x60, 0x68, 0x70, 0x78, 0x80, /* 0-4 */
71 0x88, 0x90, 0x98, 0xa0, 0xa8, /* 5-9 */
72 0xb0, 0xb8, 0xc0, 0xc8, 0xd0, /* 10-14 */
73 0xd8, 0xe0, 0xe8, 0xf0, 0xf8, /* 15-19 */
74 0x100, 0x108, 0x110, 0x118, /* 20-23 */
77 #define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0)
78 #define A7MODE(reg) GET_BITFIELD(reg, 26, 26)
80 static char *show_dram_attr(u32 attr)
82 switch (attr) {
83 case 0:
84 return "DRAM";
85 case 1:
86 return "MMCFG";
87 case 2:
88 return "NXM";
89 default:
90 return "unknown";
94 static const u32 sbridge_interleave_list[] = {
95 0x84, 0x8c, 0x94, 0x9c, 0xa4,
96 0xac, 0xb4, 0xbc, 0xc4, 0xcc,
99 static const u32 ibridge_interleave_list[] = {
100 0x64, 0x6c, 0x74, 0x7c, 0x84,
101 0x8c, 0x94, 0x9c, 0xa4, 0xac,
102 0xb4, 0xbc, 0xc4, 0xcc, 0xd4,
103 0xdc, 0xe4, 0xec, 0xf4, 0xfc,
106 static const u32 knl_interleave_list[] = {
107 0x64, 0x6c, 0x74, 0x7c, 0x84, /* 0-4 */
108 0x8c, 0x94, 0x9c, 0xa4, 0xac, /* 5-9 */
109 0xb4, 0xbc, 0xc4, 0xcc, 0xd4, /* 10-14 */
110 0xdc, 0xe4, 0xec, 0xf4, 0xfc, /* 15-19 */
111 0x104, 0x10c, 0x114, 0x11c, /* 20-23 */
113 #define MAX_INTERLEAVE \
114 (max_t(unsigned int, ARRAY_SIZE(sbridge_interleave_list), \
115 max_t(unsigned int, ARRAY_SIZE(ibridge_interleave_list), \
116 ARRAY_SIZE(knl_interleave_list))))
118 struct interleave_pkg {
119 unsigned char start;
120 unsigned char end;
123 static const struct interleave_pkg sbridge_interleave_pkg[] = {
124 { 0, 2 },
125 { 3, 5 },
126 { 8, 10 },
127 { 11, 13 },
128 { 16, 18 },
129 { 19, 21 },
130 { 24, 26 },
131 { 27, 29 },
134 static const struct interleave_pkg ibridge_interleave_pkg[] = {
135 { 0, 3 },
136 { 4, 7 },
137 { 8, 11 },
138 { 12, 15 },
139 { 16, 19 },
140 { 20, 23 },
141 { 24, 27 },
142 { 28, 31 },
145 static inline int sad_pkg(const struct interleave_pkg *table, u32 reg,
146 int interleave)
148 return GET_BITFIELD(reg, table[interleave].start,
149 table[interleave].end);
152 /* Devices 12 Function 7 */
154 #define TOLM 0x80
155 #define TOHM 0x84
156 #define HASWELL_TOLM 0xd0
157 #define HASWELL_TOHM_0 0xd4
158 #define HASWELL_TOHM_1 0xd8
159 #define KNL_TOLM 0xd0
160 #define KNL_TOHM_0 0xd4
161 #define KNL_TOHM_1 0xd8
163 #define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff)
164 #define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff)
166 /* Device 13 Function 6 */
168 #define SAD_TARGET 0xf0
170 #define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11)
172 #define SOURCE_ID_KNL(reg) GET_BITFIELD(reg, 12, 14)
174 #define SAD_CONTROL 0xf4
176 /* Device 14 function 0 */
178 static const u32 tad_dram_rule[] = {
179 0x40, 0x44, 0x48, 0x4c,
180 0x50, 0x54, 0x58, 0x5c,
181 0x60, 0x64, 0x68, 0x6c,
183 #define MAX_TAD ARRAY_SIZE(tad_dram_rule)
185 #define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff)
186 #define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11)
187 #define TAD_CH(reg) GET_BITFIELD(reg, 8, 9)
188 #define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7)
189 #define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5)
190 #define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3)
191 #define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1)
193 /* Device 15, function 0 */
195 #define MCMTR 0x7c
196 #define KNL_MCMTR 0x624
198 #define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2)
199 #define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1)
200 #define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0)
202 /* Device 15, function 1 */
204 #define RASENABLES 0xac
205 #define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0)
207 /* Device 15, functions 2-5 */
209 static const int mtr_regs[] = {
210 0x80, 0x84, 0x88,
213 static const int knl_mtr_reg = 0xb60;
215 #define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19)
216 #define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14)
217 #define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13)
218 #define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4)
219 #define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1)
221 static const u32 tad_ch_nilv_offset[] = {
222 0x90, 0x94, 0x98, 0x9c,
223 0xa0, 0xa4, 0xa8, 0xac,
224 0xb0, 0xb4, 0xb8, 0xbc,
226 #define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29)
227 #define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26)
229 static const u32 rir_way_limit[] = {
230 0x108, 0x10c, 0x110, 0x114, 0x118,
232 #define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit)
234 #define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31)
235 #define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29)
237 #define MAX_RIR_WAY 8
239 static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = {
240 { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c },
241 { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c },
242 { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c },
243 { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c },
244 { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc },
247 #define RIR_RNK_TGT(type, reg) (((type) == BROADWELL) ? \
248 GET_BITFIELD(reg, 20, 23) : GET_BITFIELD(reg, 16, 19))
250 #define RIR_OFFSET(type, reg) (((type) == HASWELL || (type) == BROADWELL) ? \
251 GET_BITFIELD(reg, 2, 15) : GET_BITFIELD(reg, 2, 14))
253 /* Device 16, functions 2-7 */
256 * FIXME: Implement the error count reads directly
259 static const u32 correrrcnt[] = {
260 0x104, 0x108, 0x10c, 0x110,
263 #define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31)
264 #define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30)
265 #define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15)
266 #define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14)
268 static const u32 correrrthrsld[] = {
269 0x11c, 0x120, 0x124, 0x128,
272 #define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30)
273 #define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14)
276 /* Device 17, function 0 */
278 #define SB_RANK_CFG_A 0x0328
280 #define IB_RANK_CFG_A 0x0320
283 * sbridge structs
286 #define NUM_CHANNELS 6 /* Max channels per MC */
287 #define MAX_DIMMS 3 /* Max DIMMS per channel */
288 #define KNL_MAX_CHAS 38 /* KNL max num. of Cache Home Agents */
289 #define KNL_MAX_CHANNELS 6 /* KNL max num. of PCI channels */
290 #define KNL_MAX_EDCS 8 /* Embedded DRAM controllers */
291 #define CHANNEL_UNSPECIFIED 0xf /* Intel IA32 SDM 15-14 */
293 enum type {
294 SANDY_BRIDGE,
295 IVY_BRIDGE,
296 HASWELL,
297 BROADWELL,
298 KNIGHTS_LANDING,
301 enum domain {
302 IMC0 = 0,
303 IMC1,
304 SOCK,
307 enum mirroring_mode {
308 NON_MIRRORING,
309 ADDR_RANGE_MIRRORING,
310 FULL_MIRRORING,
313 struct sbridge_pvt;
314 struct sbridge_info {
315 enum type type;
316 u32 mcmtr;
317 u32 rankcfgr;
318 u64 (*get_tolm)(struct sbridge_pvt *pvt);
319 u64 (*get_tohm)(struct sbridge_pvt *pvt);
320 u64 (*rir_limit)(u32 reg);
321 u64 (*sad_limit)(u32 reg);
322 u32 (*interleave_mode)(u32 reg);
323 u32 (*dram_attr)(u32 reg);
324 const u32 *dram_rule;
325 const u32 *interleave_list;
326 const struct interleave_pkg *interleave_pkg;
327 u8 max_sad;
328 u8 (*get_node_id)(struct sbridge_pvt *pvt);
329 enum mem_type (*get_memory_type)(struct sbridge_pvt *pvt);
330 enum dev_type (*get_width)(struct sbridge_pvt *pvt, u32 mtr);
331 struct pci_dev *pci_vtd;
334 struct sbridge_channel {
335 u32 ranks;
336 u32 dimms;
339 struct pci_id_descr {
340 int dev_id;
341 int optional;
342 enum domain dom;
345 struct pci_id_table {
346 const struct pci_id_descr *descr;
347 int n_devs_per_imc;
348 int n_devs_per_sock;
349 int n_imcs_per_sock;
350 enum type type;
353 struct sbridge_dev {
354 struct list_head list;
355 int seg;
356 u8 bus, mc;
357 u8 node_id, source_id;
358 struct pci_dev **pdev;
359 enum domain dom;
360 int n_devs;
361 int i_devs;
362 struct mem_ctl_info *mci;
365 struct knl_pvt {
366 struct pci_dev *pci_cha[KNL_MAX_CHAS];
367 struct pci_dev *pci_channel[KNL_MAX_CHANNELS];
368 struct pci_dev *pci_mc0;
369 struct pci_dev *pci_mc1;
370 struct pci_dev *pci_mc0_misc;
371 struct pci_dev *pci_mc1_misc;
372 struct pci_dev *pci_mc_info; /* tolm, tohm */
375 struct sbridge_pvt {
376 /* Devices per socket */
377 struct pci_dev *pci_ddrio;
378 struct pci_dev *pci_sad0, *pci_sad1;
379 struct pci_dev *pci_br0, *pci_br1;
380 /* Devices per memory controller */
381 struct pci_dev *pci_ha, *pci_ta, *pci_ras;
382 struct pci_dev *pci_tad[NUM_CHANNELS];
384 struct sbridge_dev *sbridge_dev;
386 struct sbridge_info info;
387 struct sbridge_channel channel[NUM_CHANNELS];
389 /* Memory type detection */
390 bool is_cur_addr_mirrored, is_lockstep, is_close_pg;
391 bool is_chan_hash;
392 enum mirroring_mode mirror_mode;
394 /* Memory description */
395 u64 tolm, tohm;
396 struct knl_pvt knl;
399 #define PCI_DESCR(device_id, opt, domain) \
400 .dev_id = (device_id), \
401 .optional = opt, \
402 .dom = domain
404 static const struct pci_id_descr pci_dev_descr_sbridge[] = {
405 /* Processor Home Agent */
406 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0, 0, IMC0) },
408 /* Memory controller */
409 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA, 0, IMC0) },
410 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS, 0, IMC0) },
411 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0, 0, IMC0) },
412 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1, 0, IMC0) },
413 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2, 0, IMC0) },
414 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3, 0, IMC0) },
415 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO, 1, SOCK) },
417 /* System Address Decoder */
418 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0, 0, SOCK) },
419 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1, 0, SOCK) },
421 /* Broadcast Registers */
422 { PCI_DESCR(PCI_DEVICE_ID_INTEL_SBRIDGE_BR, 0, SOCK) },
425 #define PCI_ID_TABLE_ENTRY(A, N, M, T) { \
426 .descr = A, \
427 .n_devs_per_imc = N, \
428 .n_devs_per_sock = ARRAY_SIZE(A), \
429 .n_imcs_per_sock = M, \
430 .type = T \
433 static const struct pci_id_table pci_dev_descr_sbridge_table[] = {
434 PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge, ARRAY_SIZE(pci_dev_descr_sbridge), 1, SANDY_BRIDGE),
435 {0,} /* 0 terminated list. */
438 /* This changes depending if 1HA or 2HA:
439 * 1HA:
440 * 0x0eb8 (17.0) is DDRIO0
441 * 2HA:
442 * 0x0ebc (17.4) is DDRIO0
444 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0 0x0eb8
445 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0 0x0ebc
447 /* pci ids */
448 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0 0x0ea0
449 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA 0x0ea8
450 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS 0x0e71
451 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0 0x0eaa
452 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1 0x0eab
453 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2 0x0eac
454 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3 0x0ead
455 #define PCI_DEVICE_ID_INTEL_IBRIDGE_SAD 0x0ec8
456 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR0 0x0ec9
457 #define PCI_DEVICE_ID_INTEL_IBRIDGE_BR1 0x0eca
458 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1 0x0e60
459 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA 0x0e68
460 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS 0x0e79
461 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0 0x0e6a
462 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1 0x0e6b
463 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2 0x0e6c
464 #define PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3 0x0e6d
466 static const struct pci_id_descr pci_dev_descr_ibridge[] = {
467 /* Processor Home Agent */
468 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0, 0, IMC0) },
469 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1, 1, IMC1) },
471 /* Memory controller */
472 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA, 0, IMC0) },
473 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS, 0, IMC0) },
474 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0, 0, IMC0) },
475 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1, 0, IMC0) },
476 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2, 0, IMC0) },
477 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3, 0, IMC0) },
479 /* Optional, mode 2HA */
480 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA, 1, IMC1) },
481 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS, 1, IMC1) },
482 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0, 1, IMC1) },
483 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1, 1, IMC1) },
484 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2, 1, IMC1) },
485 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3, 1, IMC1) },
487 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0, 1, SOCK) },
488 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0, 1, SOCK) },
490 /* System Address Decoder */
491 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_SAD, 0, SOCK) },
493 /* Broadcast Registers */
494 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR0, 1, SOCK) },
495 { PCI_DESCR(PCI_DEVICE_ID_INTEL_IBRIDGE_BR1, 0, SOCK) },
499 static const struct pci_id_table pci_dev_descr_ibridge_table[] = {
500 PCI_ID_TABLE_ENTRY(pci_dev_descr_ibridge, 12, 2, IVY_BRIDGE),
501 {0,} /* 0 terminated list. */
504 /* Haswell support */
505 /* EN processor:
506 * - 1 IMC
507 * - 3 DDR3 channels, 2 DPC per channel
508 * EP processor:
509 * - 1 or 2 IMC
510 * - 4 DDR4 channels, 3 DPC per channel
511 * EP 4S processor:
512 * - 2 IMC
513 * - 4 DDR4 channels, 3 DPC per channel
514 * EX processor:
515 * - 2 IMC
516 * - each IMC interfaces with a SMI 2 channel
517 * - each SMI channel interfaces with a scalable memory buffer
518 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
520 #define HASWELL_DDRCRCLKCONTROLS 0xa10 /* Ditto on Broadwell */
521 #define HASWELL_HASYSDEFEATURE2 0x84
522 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC 0x2f28
523 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0 0x2fa0
524 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1 0x2f60
525 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA 0x2fa8
526 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM 0x2f71
527 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA 0x2f68
528 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM 0x2f79
529 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0 0x2ffc
530 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1 0x2ffd
531 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0 0x2faa
532 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1 0x2fab
533 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2 0x2fac
534 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3 0x2fad
535 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0 0x2f6a
536 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1 0x2f6b
537 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2 0x2f6c
538 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3 0x2f6d
539 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0 0x2fbd
540 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1 0x2fbf
541 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2 0x2fb9
542 #define PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3 0x2fbb
543 static const struct pci_id_descr pci_dev_descr_haswell[] = {
544 /* first item must be the HA */
545 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0, 0, IMC0) },
546 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1, 1, IMC1) },
548 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA, 0, IMC0) },
549 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM, 0, IMC0) },
550 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0, 0, IMC0) },
551 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1, 0, IMC0) },
552 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2, 1, IMC0) },
553 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3, 1, IMC0) },
555 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA, 1, IMC1) },
556 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM, 1, IMC1) },
557 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0, 1, IMC1) },
558 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1, 1, IMC1) },
559 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2, 1, IMC1) },
560 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3, 1, IMC1) },
562 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0, 0, SOCK) },
563 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1, 0, SOCK) },
564 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0, 1, SOCK) },
565 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1, 1, SOCK) },
566 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2, 1, SOCK) },
567 { PCI_DESCR(PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3, 1, SOCK) },
570 static const struct pci_id_table pci_dev_descr_haswell_table[] = {
571 PCI_ID_TABLE_ENTRY(pci_dev_descr_haswell, 13, 2, HASWELL),
572 {0,} /* 0 terminated list. */
575 /* Knight's Landing Support */
577 * KNL's memory channels are swizzled between memory controllers.
578 * MC0 is mapped to CH3,4,5 and MC1 is mapped to CH0,1,2
580 #define knl_channel_remap(mc, chan) ((mc) ? (chan) : (chan) + 3)
582 /* Memory controller, TAD tables, error injection - 2-8-0, 2-9-0 (2 of these) */
583 #define PCI_DEVICE_ID_INTEL_KNL_IMC_MC 0x7840
584 /* DRAM channel stuff; bank addrs, dimmmtr, etc.. 2-8-2 - 2-9-4 (6 of these) */
585 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN 0x7843
586 /* kdrwdbu TAD limits/offsets, MCMTR - 2-10-1, 2-11-1 (2 of these) */
587 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TA 0x7844
588 /* CHA broadcast registers, dram rules - 1-29-0 (1 of these) */
589 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0 0x782a
590 /* SAD target - 1-29-1 (1 of these) */
591 #define PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1 0x782b
592 /* Caching / Home Agent */
593 #define PCI_DEVICE_ID_INTEL_KNL_IMC_CHA 0x782c
594 /* Device with TOLM and TOHM, 0-5-0 (1 of these) */
595 #define PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM 0x7810
598 * KNL differs from SB, IB, and Haswell in that it has multiple
599 * instances of the same device with the same device ID, so we handle that
600 * by creating as many copies in the table as we expect to find.
601 * (Like device ID must be grouped together.)
604 static const struct pci_id_descr pci_dev_descr_knl[] = {
605 [0 ... 1] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_MC, 0, IMC0)},
606 [2 ... 7] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN, 0, IMC0) },
607 [8] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TA, 0, IMC0) },
608 [9] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM, 0, IMC0) },
609 [10] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0, 0, SOCK) },
610 [11] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1, 0, SOCK) },
611 [12 ... 49] = { PCI_DESCR(PCI_DEVICE_ID_INTEL_KNL_IMC_CHA, 0, SOCK) },
614 static const struct pci_id_table pci_dev_descr_knl_table[] = {
615 PCI_ID_TABLE_ENTRY(pci_dev_descr_knl, ARRAY_SIZE(pci_dev_descr_knl), 1, KNIGHTS_LANDING),
616 {0,}
620 * Broadwell support
622 * DE processor:
623 * - 1 IMC
624 * - 2 DDR3 channels, 2 DPC per channel
625 * EP processor:
626 * - 1 or 2 IMC
627 * - 4 DDR4 channels, 3 DPC per channel
628 * EP 4S processor:
629 * - 2 IMC
630 * - 4 DDR4 channels, 3 DPC per channel
631 * EX processor:
632 * - 2 IMC
633 * - each IMC interfaces with a SMI 2 channel
634 * - each SMI channel interfaces with a scalable memory buffer
635 * - each scalable memory buffer supports 4 DDR3/DDR4 channels, 3 DPC
637 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC 0x6f28
638 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0 0x6fa0
639 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1 0x6f60
640 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA 0x6fa8
641 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM 0x6f71
642 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA 0x6f68
643 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM 0x6f79
644 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0 0x6ffc
645 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1 0x6ffd
646 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0 0x6faa
647 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1 0x6fab
648 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2 0x6fac
649 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3 0x6fad
650 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0 0x6f6a
651 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1 0x6f6b
652 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2 0x6f6c
653 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3 0x6f6d
654 #define PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0 0x6faf
656 static const struct pci_id_descr pci_dev_descr_broadwell[] = {
657 /* first item must be the HA */
658 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0, 0, IMC0) },
659 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1, 1, IMC1) },
661 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA, 0, IMC0) },
662 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM, 0, IMC0) },
663 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0, 0, IMC0) },
664 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1, 0, IMC0) },
665 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2, 1, IMC0) },
666 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3, 1, IMC0) },
668 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA, 1, IMC1) },
669 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM, 1, IMC1) },
670 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0, 1, IMC1) },
671 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1, 1, IMC1) },
672 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2, 1, IMC1) },
673 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3, 1, IMC1) },
675 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0, 0, SOCK) },
676 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1, 0, SOCK) },
677 { PCI_DESCR(PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0, 1, SOCK) },
680 static const struct pci_id_table pci_dev_descr_broadwell_table[] = {
681 PCI_ID_TABLE_ENTRY(pci_dev_descr_broadwell, 10, 2, BROADWELL),
682 {0,} /* 0 terminated list. */
686 /****************************************************************************
687 Ancillary status routines
688 ****************************************************************************/
690 static inline int numrank(enum type type, u32 mtr)
692 int ranks = (1 << RANK_CNT_BITS(mtr));
693 int max = 4;
695 if (type == HASWELL || type == BROADWELL || type == KNIGHTS_LANDING)
696 max = 8;
698 if (ranks > max) {
699 edac_dbg(0, "Invalid number of ranks: %d (max = %i) raw value = %x (%04x)\n",
700 ranks, max, (unsigned int)RANK_CNT_BITS(mtr), mtr);
701 return -EINVAL;
704 return ranks;
707 static inline int numrow(u32 mtr)
709 int rows = (RANK_WIDTH_BITS(mtr) + 12);
711 if (rows < 13 || rows > 18) {
712 edac_dbg(0, "Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)\n",
713 rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr);
714 return -EINVAL;
717 return 1 << rows;
720 static inline int numcol(u32 mtr)
722 int cols = (COL_WIDTH_BITS(mtr) + 10);
724 if (cols > 12) {
725 edac_dbg(0, "Invalid number of cols: %d (max = 4) raw value = %x (%04x)\n",
726 cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr);
727 return -EINVAL;
730 return 1 << cols;
733 static struct sbridge_dev *get_sbridge_dev(int seg, u8 bus, enum domain dom,
734 int multi_bus,
735 struct sbridge_dev *prev)
737 struct sbridge_dev *sbridge_dev;
740 * If we have devices scattered across several busses that pertain
741 * to the same memory controller, we'll lump them all together.
743 if (multi_bus) {
744 return list_first_entry_or_null(&sbridge_edac_list,
745 struct sbridge_dev, list);
748 sbridge_dev = list_entry(prev ? prev->list.next
749 : sbridge_edac_list.next, struct sbridge_dev, list);
751 list_for_each_entry_from(sbridge_dev, &sbridge_edac_list, list) {
752 if ((sbridge_dev->seg == seg) && (sbridge_dev->bus == bus) &&
753 (dom == SOCK || dom == sbridge_dev->dom))
754 return sbridge_dev;
757 return NULL;
760 static struct sbridge_dev *alloc_sbridge_dev(int seg, u8 bus, enum domain dom,
761 const struct pci_id_table *table)
763 struct sbridge_dev *sbridge_dev;
765 sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL);
766 if (!sbridge_dev)
767 return NULL;
769 sbridge_dev->pdev = kcalloc(table->n_devs_per_imc,
770 sizeof(*sbridge_dev->pdev),
771 GFP_KERNEL);
772 if (!sbridge_dev->pdev) {
773 kfree(sbridge_dev);
774 return NULL;
777 sbridge_dev->seg = seg;
778 sbridge_dev->bus = bus;
779 sbridge_dev->dom = dom;
780 sbridge_dev->n_devs = table->n_devs_per_imc;
781 list_add_tail(&sbridge_dev->list, &sbridge_edac_list);
783 return sbridge_dev;
786 static void free_sbridge_dev(struct sbridge_dev *sbridge_dev)
788 list_del(&sbridge_dev->list);
789 kfree(sbridge_dev->pdev);
790 kfree(sbridge_dev);
793 static u64 sbridge_get_tolm(struct sbridge_pvt *pvt)
795 u32 reg;
797 /* Address range is 32:28 */
798 pci_read_config_dword(pvt->pci_sad1, TOLM, &reg);
799 return GET_TOLM(reg);
802 static u64 sbridge_get_tohm(struct sbridge_pvt *pvt)
804 u32 reg;
806 pci_read_config_dword(pvt->pci_sad1, TOHM, &reg);
807 return GET_TOHM(reg);
810 static u64 ibridge_get_tolm(struct sbridge_pvt *pvt)
812 u32 reg;
814 pci_read_config_dword(pvt->pci_br1, TOLM, &reg);
816 return GET_TOLM(reg);
819 static u64 ibridge_get_tohm(struct sbridge_pvt *pvt)
821 u32 reg;
823 pci_read_config_dword(pvt->pci_br1, TOHM, &reg);
825 return GET_TOHM(reg);
828 static u64 rir_limit(u32 reg)
830 return ((u64)GET_BITFIELD(reg, 1, 10) << 29) | 0x1fffffff;
833 static u64 sad_limit(u32 reg)
835 return (GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff;
838 static u32 interleave_mode(u32 reg)
840 return GET_BITFIELD(reg, 1, 1);
843 static u32 dram_attr(u32 reg)
845 return GET_BITFIELD(reg, 2, 3);
848 static u64 knl_sad_limit(u32 reg)
850 return (GET_BITFIELD(reg, 7, 26) << 26) | 0x3ffffff;
853 static u32 knl_interleave_mode(u32 reg)
855 return GET_BITFIELD(reg, 1, 2);
858 static const char * const knl_intlv_mode[] = {
859 "[8:6]", "[10:8]", "[14:12]", "[32:30]"
862 static const char *get_intlv_mode_str(u32 reg, enum type t)
864 if (t == KNIGHTS_LANDING)
865 return knl_intlv_mode[knl_interleave_mode(reg)];
866 else
867 return interleave_mode(reg) ? "[8:6]" : "[8:6]XOR[18:16]";
870 static u32 dram_attr_knl(u32 reg)
872 return GET_BITFIELD(reg, 3, 4);
876 static enum mem_type get_memory_type(struct sbridge_pvt *pvt)
878 u32 reg;
879 enum mem_type mtype;
881 if (pvt->pci_ddrio) {
882 pci_read_config_dword(pvt->pci_ddrio, pvt->info.rankcfgr,
883 &reg);
884 if (GET_BITFIELD(reg, 11, 11))
885 /* FIXME: Can also be LRDIMM */
886 mtype = MEM_RDDR3;
887 else
888 mtype = MEM_DDR3;
889 } else
890 mtype = MEM_UNKNOWN;
892 return mtype;
895 static enum mem_type haswell_get_memory_type(struct sbridge_pvt *pvt)
897 u32 reg;
898 bool registered = false;
899 enum mem_type mtype = MEM_UNKNOWN;
901 if (!pvt->pci_ddrio)
902 goto out;
904 pci_read_config_dword(pvt->pci_ddrio,
905 HASWELL_DDRCRCLKCONTROLS, &reg);
906 /* Is_Rdimm */
907 if (GET_BITFIELD(reg, 16, 16))
908 registered = true;
910 pci_read_config_dword(pvt->pci_ta, MCMTR, &reg);
911 if (GET_BITFIELD(reg, 14, 14)) {
912 if (registered)
913 mtype = MEM_RDDR4;
914 else
915 mtype = MEM_DDR4;
916 } else {
917 if (registered)
918 mtype = MEM_RDDR3;
919 else
920 mtype = MEM_DDR3;
923 out:
924 return mtype;
927 static enum dev_type knl_get_width(struct sbridge_pvt *pvt, u32 mtr)
929 /* for KNL value is fixed */
930 return DEV_X16;
933 static enum dev_type sbridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
935 /* there's no way to figure out */
936 return DEV_UNKNOWN;
939 static enum dev_type __ibridge_get_width(u32 mtr)
941 enum dev_type type;
943 switch (mtr) {
944 case 3:
945 type = DEV_UNKNOWN;
946 break;
947 case 2:
948 type = DEV_X16;
949 break;
950 case 1:
951 type = DEV_X8;
952 break;
953 case 0:
954 type = DEV_X4;
955 break;
958 return type;
961 static enum dev_type ibridge_get_width(struct sbridge_pvt *pvt, u32 mtr)
964 * ddr3_width on the documentation but also valid for DDR4 on
965 * Haswell
967 return __ibridge_get_width(GET_BITFIELD(mtr, 7, 8));
970 static enum dev_type broadwell_get_width(struct sbridge_pvt *pvt, u32 mtr)
972 /* ddr3_width on the documentation but also valid for DDR4 */
973 return __ibridge_get_width(GET_BITFIELD(mtr, 8, 9));
976 static enum mem_type knl_get_memory_type(struct sbridge_pvt *pvt)
978 /* DDR4 RDIMMS and LRDIMMS are supported */
979 return MEM_RDDR4;
982 static u8 get_node_id(struct sbridge_pvt *pvt)
984 u32 reg;
985 pci_read_config_dword(pvt->pci_br0, SAD_CONTROL, &reg);
986 return GET_BITFIELD(reg, 0, 2);
989 static u8 haswell_get_node_id(struct sbridge_pvt *pvt)
991 u32 reg;
993 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, &reg);
994 return GET_BITFIELD(reg, 0, 3);
997 static u8 knl_get_node_id(struct sbridge_pvt *pvt)
999 u32 reg;
1001 pci_read_config_dword(pvt->pci_sad1, SAD_CONTROL, &reg);
1002 return GET_BITFIELD(reg, 0, 2);
1006 static u64 haswell_get_tolm(struct sbridge_pvt *pvt)
1008 u32 reg;
1010 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOLM, &reg);
1011 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1014 static u64 haswell_get_tohm(struct sbridge_pvt *pvt)
1016 u64 rc;
1017 u32 reg;
1019 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_0, &reg);
1020 rc = GET_BITFIELD(reg, 26, 31);
1021 pci_read_config_dword(pvt->info.pci_vtd, HASWELL_TOHM_1, &reg);
1022 rc = ((reg << 6) | rc) << 26;
1024 return rc | 0x1ffffff;
1027 static u64 knl_get_tolm(struct sbridge_pvt *pvt)
1029 u32 reg;
1031 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOLM, &reg);
1032 return (GET_BITFIELD(reg, 26, 31) << 26) | 0x3ffffff;
1035 static u64 knl_get_tohm(struct sbridge_pvt *pvt)
1037 u64 rc;
1038 u32 reg_lo, reg_hi;
1040 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_0, &reg_lo);
1041 pci_read_config_dword(pvt->knl.pci_mc_info, KNL_TOHM_1, &reg_hi);
1042 rc = ((u64)reg_hi << 32) | reg_lo;
1043 return rc | 0x3ffffff;
1047 static u64 haswell_rir_limit(u32 reg)
1049 return (((u64)GET_BITFIELD(reg, 1, 11) + 1) << 29) - 1;
1052 static inline u8 sad_pkg_socket(u8 pkg)
1054 /* on Ivy Bridge, nodeID is SASS, where A is HA and S is node id */
1055 return ((pkg >> 3) << 2) | (pkg & 0x3);
1058 static inline u8 sad_pkg_ha(u8 pkg)
1060 return (pkg >> 2) & 0x1;
1063 static int haswell_chan_hash(int idx, u64 addr)
1065 int i;
1068 * XOR even bits from 12:26 to bit0 of idx,
1069 * odd bits from 13:27 to bit1
1071 for (i = 12; i < 28; i += 2)
1072 idx ^= (addr >> i) & 3;
1074 return idx;
1077 /* Low bits of TAD limit, and some metadata. */
1078 static const u32 knl_tad_dram_limit_lo[] = {
1079 0x400, 0x500, 0x600, 0x700,
1080 0x800, 0x900, 0xa00, 0xb00,
1083 /* Low bits of TAD offset. */
1084 static const u32 knl_tad_dram_offset_lo[] = {
1085 0x404, 0x504, 0x604, 0x704,
1086 0x804, 0x904, 0xa04, 0xb04,
1089 /* High 16 bits of TAD limit and offset. */
1090 static const u32 knl_tad_dram_hi[] = {
1091 0x408, 0x508, 0x608, 0x708,
1092 0x808, 0x908, 0xa08, 0xb08,
1095 /* Number of ways a tad entry is interleaved. */
1096 static const u32 knl_tad_ways[] = {
1097 8, 6, 4, 3, 2, 1,
1101 * Retrieve the n'th Target Address Decode table entry
1102 * from the memory controller's TAD table.
1104 * @pvt: driver private data
1105 * @entry: which entry you want to retrieve
1106 * @mc: which memory controller (0 or 1)
1107 * @offset: output tad range offset
1108 * @limit: output address of first byte above tad range
1109 * @ways: output number of interleave ways
1111 * The offset value has curious semantics. It's a sort of running total
1112 * of the sizes of all the memory regions that aren't mapped in this
1113 * tad table.
1115 static int knl_get_tad(const struct sbridge_pvt *pvt,
1116 const int entry,
1117 const int mc,
1118 u64 *offset,
1119 u64 *limit,
1120 int *ways)
1122 u32 reg_limit_lo, reg_offset_lo, reg_hi;
1123 struct pci_dev *pci_mc;
1124 int way_id;
1126 switch (mc) {
1127 case 0:
1128 pci_mc = pvt->knl.pci_mc0;
1129 break;
1130 case 1:
1131 pci_mc = pvt->knl.pci_mc1;
1132 break;
1133 default:
1134 WARN_ON(1);
1135 return -EINVAL;
1138 pci_read_config_dword(pci_mc,
1139 knl_tad_dram_limit_lo[entry], &reg_limit_lo);
1140 pci_read_config_dword(pci_mc,
1141 knl_tad_dram_offset_lo[entry], &reg_offset_lo);
1142 pci_read_config_dword(pci_mc,
1143 knl_tad_dram_hi[entry], &reg_hi);
1145 /* Is this TAD entry enabled? */
1146 if (!GET_BITFIELD(reg_limit_lo, 0, 0))
1147 return -ENODEV;
1149 way_id = GET_BITFIELD(reg_limit_lo, 3, 5);
1151 if (way_id < ARRAY_SIZE(knl_tad_ways)) {
1152 *ways = knl_tad_ways[way_id];
1153 } else {
1154 *ways = 0;
1155 sbridge_printk(KERN_ERR,
1156 "Unexpected value %d in mc_tad_limit_lo wayness field\n",
1157 way_id);
1158 return -ENODEV;
1162 * The least significant 6 bits of base and limit are truncated.
1163 * For limit, we fill the missing bits with 1s.
1165 *offset = ((u64) GET_BITFIELD(reg_offset_lo, 6, 31) << 6) |
1166 ((u64) GET_BITFIELD(reg_hi, 0, 15) << 32);
1167 *limit = ((u64) GET_BITFIELD(reg_limit_lo, 6, 31) << 6) | 63 |
1168 ((u64) GET_BITFIELD(reg_hi, 16, 31) << 32);
1170 return 0;
1173 /* Determine which memory controller is responsible for a given channel. */
1174 static int knl_channel_mc(int channel)
1176 WARN_ON(channel < 0 || channel >= 6);
1178 return channel < 3 ? 1 : 0;
1182 * Get the Nth entry from EDC_ROUTE_TABLE register.
1183 * (This is the per-tile mapping of logical interleave targets to
1184 * physical EDC modules.)
1186 * entry 0: 0:2
1187 * 1: 3:5
1188 * 2: 6:8
1189 * 3: 9:11
1190 * 4: 12:14
1191 * 5: 15:17
1192 * 6: 18:20
1193 * 7: 21:23
1194 * reserved: 24:31
1196 static u32 knl_get_edc_route(int entry, u32 reg)
1198 WARN_ON(entry >= KNL_MAX_EDCS);
1199 return GET_BITFIELD(reg, entry*3, (entry*3)+2);
1203 * Get the Nth entry from MC_ROUTE_TABLE register.
1204 * (This is the per-tile mapping of logical interleave targets to
1205 * physical DRAM channels modules.)
1207 * entry 0: mc 0:2 channel 18:19
1208 * 1: mc 3:5 channel 20:21
1209 * 2: mc 6:8 channel 22:23
1210 * 3: mc 9:11 channel 24:25
1211 * 4: mc 12:14 channel 26:27
1212 * 5: mc 15:17 channel 28:29
1213 * reserved: 30:31
1215 * Though we have 3 bits to identify the MC, we should only see
1216 * the values 0 or 1.
1219 static u32 knl_get_mc_route(int entry, u32 reg)
1221 int mc, chan;
1223 WARN_ON(entry >= KNL_MAX_CHANNELS);
1225 mc = GET_BITFIELD(reg, entry*3, (entry*3)+2);
1226 chan = GET_BITFIELD(reg, (entry*2) + 18, (entry*2) + 18 + 1);
1228 return knl_channel_remap(mc, chan);
1232 * Render the EDC_ROUTE register in human-readable form.
1233 * Output string s should be at least KNL_MAX_EDCS*2 bytes.
1235 static void knl_show_edc_route(u32 reg, char *s)
1237 int i;
1239 for (i = 0; i < KNL_MAX_EDCS; i++) {
1240 s[i*2] = knl_get_edc_route(i, reg) + '0';
1241 s[i*2+1] = '-';
1244 s[KNL_MAX_EDCS*2 - 1] = '\0';
1248 * Render the MC_ROUTE register in human-readable form.
1249 * Output string s should be at least KNL_MAX_CHANNELS*2 bytes.
1251 static void knl_show_mc_route(u32 reg, char *s)
1253 int i;
1255 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
1256 s[i*2] = knl_get_mc_route(i, reg) + '0';
1257 s[i*2+1] = '-';
1260 s[KNL_MAX_CHANNELS*2 - 1] = '\0';
1263 #define KNL_EDC_ROUTE 0xb8
1264 #define KNL_MC_ROUTE 0xb4
1266 /* Is this dram rule backed by regular DRAM in flat mode? */
1267 #define KNL_EDRAM(reg) GET_BITFIELD(reg, 29, 29)
1269 /* Is this dram rule cached? */
1270 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1272 /* Is this rule backed by edc ? */
1273 #define KNL_EDRAM_ONLY(reg) GET_BITFIELD(reg, 29, 29)
1275 /* Is this rule backed by DRAM, cacheable in EDRAM? */
1276 #define KNL_CACHEABLE(reg) GET_BITFIELD(reg, 28, 28)
1278 /* Is this rule mod3? */
1279 #define KNL_MOD3(reg) GET_BITFIELD(reg, 27, 27)
1282 * Figure out how big our RAM modules are.
1284 * The DIMMMTR register in KNL doesn't tell us the size of the DIMMs, so we
1285 * have to figure this out from the SAD rules, interleave lists, route tables,
1286 * and TAD rules.
1288 * SAD rules can have holes in them (e.g. the 3G-4G hole), so we have to
1289 * inspect the TAD rules to figure out how large the SAD regions really are.
1291 * When we know the real size of a SAD region and how many ways it's
1292 * interleaved, we know the individual contribution of each channel to
1293 * TAD is size/ways.
1295 * Finally, we have to check whether each channel participates in each SAD
1296 * region.
1298 * Fortunately, KNL only supports one DIMM per channel, so once we know how
1299 * much memory the channel uses, we know the DIMM is at least that large.
1300 * (The BIOS might possibly choose not to map all available memory, in which
1301 * case we will underreport the size of the DIMM.)
1303 * In theory, we could try to determine the EDC sizes as well, but that would
1304 * only work in flat mode, not in cache mode.
1306 * @mc_sizes: Output sizes of channels (must have space for KNL_MAX_CHANNELS
1307 * elements)
1309 static int knl_get_dimm_capacity(struct sbridge_pvt *pvt, u64 *mc_sizes)
1311 u64 sad_base, sad_size, sad_limit = 0;
1312 u64 tad_base, tad_size, tad_limit, tad_deadspace, tad_livespace;
1313 int sad_rule = 0;
1314 int tad_rule = 0;
1315 int intrlv_ways, tad_ways;
1316 u32 first_pkg, pkg;
1317 int i;
1318 u64 sad_actual_size[2]; /* sad size accounting for holes, per mc */
1319 u32 dram_rule, interleave_reg;
1320 u32 mc_route_reg[KNL_MAX_CHAS];
1321 u32 edc_route_reg[KNL_MAX_CHAS];
1322 int edram_only;
1323 char edc_route_string[KNL_MAX_EDCS*2];
1324 char mc_route_string[KNL_MAX_CHANNELS*2];
1325 int cur_reg_start;
1326 int mc;
1327 int channel;
1328 int participants[KNL_MAX_CHANNELS];
1330 for (i = 0; i < KNL_MAX_CHANNELS; i++)
1331 mc_sizes[i] = 0;
1333 /* Read the EDC route table in each CHA. */
1334 cur_reg_start = 0;
1335 for (i = 0; i < KNL_MAX_CHAS; i++) {
1336 pci_read_config_dword(pvt->knl.pci_cha[i],
1337 KNL_EDC_ROUTE, &edc_route_reg[i]);
1339 if (i > 0 && edc_route_reg[i] != edc_route_reg[i-1]) {
1340 knl_show_edc_route(edc_route_reg[i-1],
1341 edc_route_string);
1342 if (cur_reg_start == i-1)
1343 edac_dbg(0, "edc route table for CHA %d: %s\n",
1344 cur_reg_start, edc_route_string);
1345 else
1346 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1347 cur_reg_start, i-1, edc_route_string);
1348 cur_reg_start = i;
1351 knl_show_edc_route(edc_route_reg[i-1], edc_route_string);
1352 if (cur_reg_start == i-1)
1353 edac_dbg(0, "edc route table for CHA %d: %s\n",
1354 cur_reg_start, edc_route_string);
1355 else
1356 edac_dbg(0, "edc route table for CHA %d-%d: %s\n",
1357 cur_reg_start, i-1, edc_route_string);
1359 /* Read the MC route table in each CHA. */
1360 cur_reg_start = 0;
1361 for (i = 0; i < KNL_MAX_CHAS; i++) {
1362 pci_read_config_dword(pvt->knl.pci_cha[i],
1363 KNL_MC_ROUTE, &mc_route_reg[i]);
1365 if (i > 0 && mc_route_reg[i] != mc_route_reg[i-1]) {
1366 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1367 if (cur_reg_start == i-1)
1368 edac_dbg(0, "mc route table for CHA %d: %s\n",
1369 cur_reg_start, mc_route_string);
1370 else
1371 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1372 cur_reg_start, i-1, mc_route_string);
1373 cur_reg_start = i;
1376 knl_show_mc_route(mc_route_reg[i-1], mc_route_string);
1377 if (cur_reg_start == i-1)
1378 edac_dbg(0, "mc route table for CHA %d: %s\n",
1379 cur_reg_start, mc_route_string);
1380 else
1381 edac_dbg(0, "mc route table for CHA %d-%d: %s\n",
1382 cur_reg_start, i-1, mc_route_string);
1384 /* Process DRAM rules */
1385 for (sad_rule = 0; sad_rule < pvt->info.max_sad; sad_rule++) {
1386 /* previous limit becomes the new base */
1387 sad_base = sad_limit;
1389 pci_read_config_dword(pvt->pci_sad0,
1390 pvt->info.dram_rule[sad_rule], &dram_rule);
1392 if (!DRAM_RULE_ENABLE(dram_rule))
1393 break;
1395 edram_only = KNL_EDRAM_ONLY(dram_rule);
1397 sad_limit = pvt->info.sad_limit(dram_rule)+1;
1398 sad_size = sad_limit - sad_base;
1400 pci_read_config_dword(pvt->pci_sad0,
1401 pvt->info.interleave_list[sad_rule], &interleave_reg);
1404 * Find out how many ways this dram rule is interleaved.
1405 * We stop when we see the first channel again.
1407 first_pkg = sad_pkg(pvt->info.interleave_pkg,
1408 interleave_reg, 0);
1409 for (intrlv_ways = 1; intrlv_ways < 8; intrlv_ways++) {
1410 pkg = sad_pkg(pvt->info.interleave_pkg,
1411 interleave_reg, intrlv_ways);
1413 if ((pkg & 0x8) == 0) {
1415 * 0 bit means memory is non-local,
1416 * which KNL doesn't support
1418 edac_dbg(0, "Unexpected interleave target %d\n",
1419 pkg);
1420 return -1;
1423 if (pkg == first_pkg)
1424 break;
1426 if (KNL_MOD3(dram_rule))
1427 intrlv_ways *= 3;
1429 edac_dbg(3, "dram rule %d (base 0x%llx, limit 0x%llx), %d way interleave%s\n",
1430 sad_rule,
1431 sad_base,
1432 sad_limit,
1433 intrlv_ways,
1434 edram_only ? ", EDRAM" : "");
1437 * Find out how big the SAD region really is by iterating
1438 * over TAD tables (SAD regions may contain holes).
1439 * Each memory controller might have a different TAD table, so
1440 * we have to look at both.
1442 * Livespace is the memory that's mapped in this TAD table,
1443 * deadspace is the holes (this could be the MMIO hole, or it
1444 * could be memory that's mapped by the other TAD table but
1445 * not this one).
1447 for (mc = 0; mc < 2; mc++) {
1448 sad_actual_size[mc] = 0;
1449 tad_livespace = 0;
1450 for (tad_rule = 0;
1451 tad_rule < ARRAY_SIZE(
1452 knl_tad_dram_limit_lo);
1453 tad_rule++) {
1454 if (knl_get_tad(pvt,
1455 tad_rule,
1457 &tad_deadspace,
1458 &tad_limit,
1459 &tad_ways))
1460 break;
1462 tad_size = (tad_limit+1) -
1463 (tad_livespace + tad_deadspace);
1464 tad_livespace += tad_size;
1465 tad_base = (tad_limit+1) - tad_size;
1467 if (tad_base < sad_base) {
1468 if (tad_limit > sad_base)
1469 edac_dbg(0, "TAD region overlaps lower SAD boundary -- TAD tables may be configured incorrectly.\n");
1470 } else if (tad_base < sad_limit) {
1471 if (tad_limit+1 > sad_limit) {
1472 edac_dbg(0, "TAD region overlaps upper SAD boundary -- TAD tables may be configured incorrectly.\n");
1473 } else {
1474 /* TAD region is completely inside SAD region */
1475 edac_dbg(3, "TAD region %d 0x%llx - 0x%llx (%lld bytes) table%d\n",
1476 tad_rule, tad_base,
1477 tad_limit, tad_size,
1478 mc);
1479 sad_actual_size[mc] += tad_size;
1482 tad_base = tad_limit+1;
1486 for (mc = 0; mc < 2; mc++) {
1487 edac_dbg(3, " total TAD DRAM footprint in table%d : 0x%llx (%lld bytes)\n",
1488 mc, sad_actual_size[mc], sad_actual_size[mc]);
1491 /* Ignore EDRAM rule */
1492 if (edram_only)
1493 continue;
1495 /* Figure out which channels participate in interleave. */
1496 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++)
1497 participants[channel] = 0;
1499 /* For each channel, does at least one CHA have
1500 * this channel mapped to the given target?
1502 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1503 int target;
1504 int cha;
1506 for (target = 0; target < KNL_MAX_CHANNELS; target++) {
1507 for (cha = 0; cha < KNL_MAX_CHAS; cha++) {
1508 if (knl_get_mc_route(target,
1509 mc_route_reg[cha]) == channel
1510 && !participants[channel]) {
1511 participants[channel] = 1;
1512 break;
1518 for (channel = 0; channel < KNL_MAX_CHANNELS; channel++) {
1519 mc = knl_channel_mc(channel);
1520 if (participants[channel]) {
1521 edac_dbg(4, "mc channel %d contributes %lld bytes via sad entry %d\n",
1522 channel,
1523 sad_actual_size[mc]/intrlv_ways,
1524 sad_rule);
1525 mc_sizes[channel] +=
1526 sad_actual_size[mc]/intrlv_ways;
1531 return 0;
1534 static void get_source_id(struct mem_ctl_info *mci)
1536 struct sbridge_pvt *pvt = mci->pvt_info;
1537 u32 reg;
1539 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL ||
1540 pvt->info.type == KNIGHTS_LANDING)
1541 pci_read_config_dword(pvt->pci_sad1, SAD_TARGET, &reg);
1542 else
1543 pci_read_config_dword(pvt->pci_br0, SAD_TARGET, &reg);
1545 if (pvt->info.type == KNIGHTS_LANDING)
1546 pvt->sbridge_dev->source_id = SOURCE_ID_KNL(reg);
1547 else
1548 pvt->sbridge_dev->source_id = SOURCE_ID(reg);
1551 static int __populate_dimms(struct mem_ctl_info *mci,
1552 u64 knl_mc_sizes[KNL_MAX_CHANNELS],
1553 enum edac_type mode)
1555 struct sbridge_pvt *pvt = mci->pvt_info;
1556 int channels = pvt->info.type == KNIGHTS_LANDING ? KNL_MAX_CHANNELS
1557 : NUM_CHANNELS;
1558 unsigned int i, j, banks, ranks, rows, cols, npages;
1559 struct dimm_info *dimm;
1560 enum mem_type mtype;
1561 u64 size;
1563 mtype = pvt->info.get_memory_type(pvt);
1564 if (mtype == MEM_RDDR3 || mtype == MEM_RDDR4)
1565 edac_dbg(0, "Memory is registered\n");
1566 else if (mtype == MEM_UNKNOWN)
1567 edac_dbg(0, "Cannot determine memory type\n");
1568 else
1569 edac_dbg(0, "Memory is unregistered\n");
1571 if (mtype == MEM_DDR4 || mtype == MEM_RDDR4)
1572 banks = 16;
1573 else
1574 banks = 8;
1576 for (i = 0; i < channels; i++) {
1577 u32 mtr;
1579 int max_dimms_per_channel;
1581 if (pvt->info.type == KNIGHTS_LANDING) {
1582 max_dimms_per_channel = 1;
1583 if (!pvt->knl.pci_channel[i])
1584 continue;
1585 } else {
1586 max_dimms_per_channel = ARRAY_SIZE(mtr_regs);
1587 if (!pvt->pci_tad[i])
1588 continue;
1591 for (j = 0; j < max_dimms_per_channel; j++) {
1592 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms, mci->n_layers, i, j, 0);
1593 if (pvt->info.type == KNIGHTS_LANDING) {
1594 pci_read_config_dword(pvt->knl.pci_channel[i],
1595 knl_mtr_reg, &mtr);
1596 } else {
1597 pci_read_config_dword(pvt->pci_tad[i],
1598 mtr_regs[j], &mtr);
1600 edac_dbg(4, "Channel #%d MTR%d = %x\n", i, j, mtr);
1601 if (IS_DIMM_PRESENT(mtr)) {
1602 if (!IS_ECC_ENABLED(pvt->info.mcmtr)) {
1603 sbridge_printk(KERN_ERR, "CPU SrcID #%d, Ha #%d, Channel #%d has DIMMs, but ECC is disabled\n",
1604 pvt->sbridge_dev->source_id,
1605 pvt->sbridge_dev->dom, i);
1606 return -ENODEV;
1608 pvt->channel[i].dimms++;
1610 ranks = numrank(pvt->info.type, mtr);
1612 if (pvt->info.type == KNIGHTS_LANDING) {
1613 /* For DDR4, this is fixed. */
1614 cols = 1 << 10;
1615 rows = knl_mc_sizes[i] /
1616 ((u64) cols * ranks * banks * 8);
1617 } else {
1618 rows = numrow(mtr);
1619 cols = numcol(mtr);
1622 size = ((u64)rows * cols * banks * ranks) >> (20 - 3);
1623 npages = MiB_TO_PAGES(size);
1625 edac_dbg(0, "mc#%d: ha %d channel %d, dimm %d, %lld Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n",
1626 pvt->sbridge_dev->mc, pvt->sbridge_dev->dom, i, j,
1627 size, npages,
1628 banks, ranks, rows, cols);
1630 dimm->nr_pages = npages;
1631 dimm->grain = 32;
1632 dimm->dtype = pvt->info.get_width(pvt, mtr);
1633 dimm->mtype = mtype;
1634 dimm->edac_mode = mode;
1635 snprintf(dimm->label, sizeof(dimm->label),
1636 "CPU_SrcID#%u_Ha#%u_Chan#%u_DIMM#%u",
1637 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom, i, j);
1642 return 0;
1645 static int get_dimm_config(struct mem_ctl_info *mci)
1647 struct sbridge_pvt *pvt = mci->pvt_info;
1648 u64 knl_mc_sizes[KNL_MAX_CHANNELS];
1649 enum edac_type mode;
1650 u32 reg;
1652 pvt->sbridge_dev->node_id = pvt->info.get_node_id(pvt);
1653 edac_dbg(0, "mc#%d: Node ID: %d, source ID: %d\n",
1654 pvt->sbridge_dev->mc,
1655 pvt->sbridge_dev->node_id,
1656 pvt->sbridge_dev->source_id);
1658 /* KNL doesn't support mirroring or lockstep,
1659 * and is always closed page
1661 if (pvt->info.type == KNIGHTS_LANDING) {
1662 mode = EDAC_S4ECD4ED;
1663 pvt->mirror_mode = NON_MIRRORING;
1664 pvt->is_cur_addr_mirrored = false;
1666 if (knl_get_dimm_capacity(pvt, knl_mc_sizes) != 0)
1667 return -1;
1668 if (pci_read_config_dword(pvt->pci_ta, KNL_MCMTR, &pvt->info.mcmtr)) {
1669 edac_dbg(0, "Failed to read KNL_MCMTR register\n");
1670 return -ENODEV;
1672 } else {
1673 if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
1674 if (pci_read_config_dword(pvt->pci_ha, HASWELL_HASYSDEFEATURE2, &reg)) {
1675 edac_dbg(0, "Failed to read HASWELL_HASYSDEFEATURE2 register\n");
1676 return -ENODEV;
1678 pvt->is_chan_hash = GET_BITFIELD(reg, 21, 21);
1679 if (GET_BITFIELD(reg, 28, 28)) {
1680 pvt->mirror_mode = ADDR_RANGE_MIRRORING;
1681 edac_dbg(0, "Address range partial memory mirroring is enabled\n");
1682 goto next;
1685 if (pci_read_config_dword(pvt->pci_ras, RASENABLES, &reg)) {
1686 edac_dbg(0, "Failed to read RASENABLES register\n");
1687 return -ENODEV;
1689 if (IS_MIRROR_ENABLED(reg)) {
1690 pvt->mirror_mode = FULL_MIRRORING;
1691 edac_dbg(0, "Full memory mirroring is enabled\n");
1692 } else {
1693 pvt->mirror_mode = NON_MIRRORING;
1694 edac_dbg(0, "Memory mirroring is disabled\n");
1697 next:
1698 if (pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr)) {
1699 edac_dbg(0, "Failed to read MCMTR register\n");
1700 return -ENODEV;
1702 if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) {
1703 edac_dbg(0, "Lockstep is enabled\n");
1704 mode = EDAC_S8ECD8ED;
1705 pvt->is_lockstep = true;
1706 } else {
1707 edac_dbg(0, "Lockstep is disabled\n");
1708 mode = EDAC_S4ECD4ED;
1709 pvt->is_lockstep = false;
1711 if (IS_CLOSE_PG(pvt->info.mcmtr)) {
1712 edac_dbg(0, "address map is on closed page mode\n");
1713 pvt->is_close_pg = true;
1714 } else {
1715 edac_dbg(0, "address map is on open page mode\n");
1716 pvt->is_close_pg = false;
1720 return __populate_dimms(mci, knl_mc_sizes, mode);
1723 static void get_memory_layout(const struct mem_ctl_info *mci)
1725 struct sbridge_pvt *pvt = mci->pvt_info;
1726 int i, j, k, n_sads, n_tads, sad_interl;
1727 u32 reg;
1728 u64 limit, prv = 0;
1729 u64 tmp_mb;
1730 u32 gb, mb;
1731 u32 rir_way;
1734 * Step 1) Get TOLM/TOHM ranges
1737 pvt->tolm = pvt->info.get_tolm(pvt);
1738 tmp_mb = (1 + pvt->tolm) >> 20;
1740 gb = div_u64_rem(tmp_mb, 1024, &mb);
1741 edac_dbg(0, "TOLM: %u.%03u GB (0x%016Lx)\n",
1742 gb, (mb*1000)/1024, (u64)pvt->tolm);
1744 /* Address range is already 45:25 */
1745 pvt->tohm = pvt->info.get_tohm(pvt);
1746 tmp_mb = (1 + pvt->tohm) >> 20;
1748 gb = div_u64_rem(tmp_mb, 1024, &mb);
1749 edac_dbg(0, "TOHM: %u.%03u GB (0x%016Lx)\n",
1750 gb, (mb*1000)/1024, (u64)pvt->tohm);
1753 * Step 2) Get SAD range and SAD Interleave list
1754 * TAD registers contain the interleave wayness. However, it
1755 * seems simpler to just discover it indirectly, with the
1756 * algorithm bellow.
1758 prv = 0;
1759 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1760 /* SAD_LIMIT Address range is 45:26 */
1761 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1762 &reg);
1763 limit = pvt->info.sad_limit(reg);
1765 if (!DRAM_RULE_ENABLE(reg))
1766 continue;
1768 if (limit <= prv)
1769 break;
1771 tmp_mb = (limit + 1) >> 20;
1772 gb = div_u64_rem(tmp_mb, 1024, &mb);
1773 edac_dbg(0, "SAD#%d %s up to %u.%03u GB (0x%016Lx) Interleave: %s reg=0x%08x\n",
1774 n_sads,
1775 show_dram_attr(pvt->info.dram_attr(reg)),
1776 gb, (mb*1000)/1024,
1777 ((u64)tmp_mb) << 20L,
1778 get_intlv_mode_str(reg, pvt->info.type),
1779 reg);
1780 prv = limit;
1782 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1783 &reg);
1784 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1785 for (j = 0; j < 8; j++) {
1786 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, j);
1787 if (j > 0 && sad_interl == pkg)
1788 break;
1790 edac_dbg(0, "SAD#%d, interleave #%d: %d\n",
1791 n_sads, j, pkg);
1795 if (pvt->info.type == KNIGHTS_LANDING)
1796 return;
1799 * Step 3) Get TAD range
1801 prv = 0;
1802 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
1803 pci_read_config_dword(pvt->pci_ha, tad_dram_rule[n_tads], &reg);
1804 limit = TAD_LIMIT(reg);
1805 if (limit <= prv)
1806 break;
1807 tmp_mb = (limit + 1) >> 20;
1809 gb = div_u64_rem(tmp_mb, 1024, &mb);
1810 edac_dbg(0, "TAD#%d: up to %u.%03u GB (0x%016Lx), socket interleave %d, memory interleave %d, TGT: %d, %d, %d, %d, reg=0x%08x\n",
1811 n_tads, gb, (mb*1000)/1024,
1812 ((u64)tmp_mb) << 20L,
1813 (u32)(1 << TAD_SOCK(reg)),
1814 (u32)TAD_CH(reg) + 1,
1815 (u32)TAD_TGT0(reg),
1816 (u32)TAD_TGT1(reg),
1817 (u32)TAD_TGT2(reg),
1818 (u32)TAD_TGT3(reg),
1819 reg);
1820 prv = limit;
1824 * Step 4) Get TAD offsets, per each channel
1826 for (i = 0; i < NUM_CHANNELS; i++) {
1827 if (!pvt->channel[i].dimms)
1828 continue;
1829 for (j = 0; j < n_tads; j++) {
1830 pci_read_config_dword(pvt->pci_tad[i],
1831 tad_ch_nilv_offset[j],
1832 &reg);
1833 tmp_mb = TAD_OFFSET(reg) >> 20;
1834 gb = div_u64_rem(tmp_mb, 1024, &mb);
1835 edac_dbg(0, "TAD CH#%d, offset #%d: %u.%03u GB (0x%016Lx), reg=0x%08x\n",
1836 i, j,
1837 gb, (mb*1000)/1024,
1838 ((u64)tmp_mb) << 20L,
1839 reg);
1844 * Step 6) Get RIR Wayness/Limit, per each channel
1846 for (i = 0; i < NUM_CHANNELS; i++) {
1847 if (!pvt->channel[i].dimms)
1848 continue;
1849 for (j = 0; j < MAX_RIR_RANGES; j++) {
1850 pci_read_config_dword(pvt->pci_tad[i],
1851 rir_way_limit[j],
1852 &reg);
1854 if (!IS_RIR_VALID(reg))
1855 continue;
1857 tmp_mb = pvt->info.rir_limit(reg) >> 20;
1858 rir_way = 1 << RIR_WAY(reg);
1859 gb = div_u64_rem(tmp_mb, 1024, &mb);
1860 edac_dbg(0, "CH#%d RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d, reg=0x%08x\n",
1861 i, j,
1862 gb, (mb*1000)/1024,
1863 ((u64)tmp_mb) << 20L,
1864 rir_way,
1865 reg);
1867 for (k = 0; k < rir_way; k++) {
1868 pci_read_config_dword(pvt->pci_tad[i],
1869 rir_offset[j][k],
1870 &reg);
1871 tmp_mb = RIR_OFFSET(pvt->info.type, reg) << 6;
1873 gb = div_u64_rem(tmp_mb, 1024, &mb);
1874 edac_dbg(0, "CH#%d RIR#%d INTL#%d, offset %u.%03u GB (0x%016Lx), tgt: %d, reg=0x%08x\n",
1875 i, j, k,
1876 gb, (mb*1000)/1024,
1877 ((u64)tmp_mb) << 20L,
1878 (u32)RIR_RNK_TGT(pvt->info.type, reg),
1879 reg);
1885 static struct mem_ctl_info *get_mci_for_node_id(u8 node_id, u8 ha)
1887 struct sbridge_dev *sbridge_dev;
1889 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
1890 if (sbridge_dev->node_id == node_id && sbridge_dev->dom == ha)
1891 return sbridge_dev->mci;
1893 return NULL;
1896 static int get_memory_error_data(struct mem_ctl_info *mci,
1897 u64 addr,
1898 u8 *socket, u8 *ha,
1899 long *channel_mask,
1900 u8 *rank,
1901 char **area_type, char *msg)
1903 struct mem_ctl_info *new_mci;
1904 struct sbridge_pvt *pvt = mci->pvt_info;
1905 struct pci_dev *pci_ha;
1906 int n_rir, n_sads, n_tads, sad_way, sck_xch;
1907 int sad_interl, idx, base_ch;
1908 int interleave_mode, shiftup = 0;
1909 unsigned int sad_interleave[MAX_INTERLEAVE];
1910 u32 reg, dram_rule;
1911 u8 ch_way, sck_way, pkg, sad_ha = 0;
1912 u32 tad_offset;
1913 u32 rir_way;
1914 u32 mb, gb;
1915 u64 ch_addr, offset, limit = 0, prv = 0;
1919 * Step 0) Check if the address is at special memory ranges
1920 * The check bellow is probably enough to fill all cases where
1921 * the error is not inside a memory, except for the legacy
1922 * range (e. g. VGA addresses). It is unlikely, however, that the
1923 * memory controller would generate an error on that range.
1925 if ((addr > (u64) pvt->tolm) && (addr < (1LL << 32))) {
1926 sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr);
1927 return -EINVAL;
1929 if (addr >= (u64)pvt->tohm) {
1930 sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr);
1931 return -EINVAL;
1935 * Step 1) Get socket
1937 for (n_sads = 0; n_sads < pvt->info.max_sad; n_sads++) {
1938 pci_read_config_dword(pvt->pci_sad0, pvt->info.dram_rule[n_sads],
1939 &reg);
1941 if (!DRAM_RULE_ENABLE(reg))
1942 continue;
1944 limit = pvt->info.sad_limit(reg);
1945 if (limit <= prv) {
1946 sprintf(msg, "Can't discover the memory socket");
1947 return -EINVAL;
1949 if (addr <= limit)
1950 break;
1951 prv = limit;
1953 if (n_sads == pvt->info.max_sad) {
1954 sprintf(msg, "Can't discover the memory socket");
1955 return -EINVAL;
1957 dram_rule = reg;
1958 *area_type = show_dram_attr(pvt->info.dram_attr(dram_rule));
1959 interleave_mode = pvt->info.interleave_mode(dram_rule);
1961 pci_read_config_dword(pvt->pci_sad0, pvt->info.interleave_list[n_sads],
1962 &reg);
1964 if (pvt->info.type == SANDY_BRIDGE) {
1965 sad_interl = sad_pkg(pvt->info.interleave_pkg, reg, 0);
1966 for (sad_way = 0; sad_way < 8; sad_way++) {
1967 u32 pkg = sad_pkg(pvt->info.interleave_pkg, reg, sad_way);
1968 if (sad_way > 0 && sad_interl == pkg)
1969 break;
1970 sad_interleave[sad_way] = pkg;
1971 edac_dbg(0, "SAD interleave #%d: %d\n",
1972 sad_way, sad_interleave[sad_way]);
1974 edac_dbg(0, "mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n",
1975 pvt->sbridge_dev->mc,
1976 n_sads,
1977 addr,
1978 limit,
1979 sad_way + 7,
1980 !interleave_mode ? "" : "XOR[18:16]");
1981 if (interleave_mode)
1982 idx = ((addr >> 6) ^ (addr >> 16)) & 7;
1983 else
1984 idx = (addr >> 6) & 7;
1985 switch (sad_way) {
1986 case 1:
1987 idx = 0;
1988 break;
1989 case 2:
1990 idx = idx & 1;
1991 break;
1992 case 4:
1993 idx = idx & 3;
1994 break;
1995 case 8:
1996 break;
1997 default:
1998 sprintf(msg, "Can't discover socket interleave");
1999 return -EINVAL;
2001 *socket = sad_interleave[idx];
2002 edac_dbg(0, "SAD interleave index: %d (wayness %d) = CPU socket %d\n",
2003 idx, sad_way, *socket);
2004 } else if (pvt->info.type == HASWELL || pvt->info.type == BROADWELL) {
2005 int bits, a7mode = A7MODE(dram_rule);
2007 if (a7mode) {
2008 /* A7 mode swaps P9 with P6 */
2009 bits = GET_BITFIELD(addr, 7, 8) << 1;
2010 bits |= GET_BITFIELD(addr, 9, 9);
2011 } else
2012 bits = GET_BITFIELD(addr, 6, 8);
2014 if (interleave_mode == 0) {
2015 /* interleave mode will XOR {8,7,6} with {18,17,16} */
2016 idx = GET_BITFIELD(addr, 16, 18);
2017 idx ^= bits;
2018 } else
2019 idx = bits;
2021 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2022 *socket = sad_pkg_socket(pkg);
2023 sad_ha = sad_pkg_ha(pkg);
2025 if (a7mode) {
2026 /* MCChanShiftUpEnable */
2027 pci_read_config_dword(pvt->pci_ha, HASWELL_HASYSDEFEATURE2, &reg);
2028 shiftup = GET_BITFIELD(reg, 22, 22);
2031 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %i, shiftup: %i\n",
2032 idx, *socket, sad_ha, shiftup);
2033 } else {
2034 /* Ivy Bridge's SAD mode doesn't support XOR interleave mode */
2035 idx = (addr >> 6) & 7;
2036 pkg = sad_pkg(pvt->info.interleave_pkg, reg, idx);
2037 *socket = sad_pkg_socket(pkg);
2038 sad_ha = sad_pkg_ha(pkg);
2039 edac_dbg(0, "SAD interleave package: %d = CPU socket %d, HA %d\n",
2040 idx, *socket, sad_ha);
2043 *ha = sad_ha;
2046 * Move to the proper node structure, in order to access the
2047 * right PCI registers
2049 new_mci = get_mci_for_node_id(*socket, sad_ha);
2050 if (!new_mci) {
2051 sprintf(msg, "Struct for socket #%u wasn't initialized",
2052 *socket);
2053 return -EINVAL;
2055 mci = new_mci;
2056 pvt = mci->pvt_info;
2059 * Step 2) Get memory channel
2061 prv = 0;
2062 pci_ha = pvt->pci_ha;
2063 for (n_tads = 0; n_tads < MAX_TAD; n_tads++) {
2064 pci_read_config_dword(pci_ha, tad_dram_rule[n_tads], &reg);
2065 limit = TAD_LIMIT(reg);
2066 if (limit <= prv) {
2067 sprintf(msg, "Can't discover the memory channel");
2068 return -EINVAL;
2070 if (addr <= limit)
2071 break;
2072 prv = limit;
2074 if (n_tads == MAX_TAD) {
2075 sprintf(msg, "Can't discover the memory channel");
2076 return -EINVAL;
2079 ch_way = TAD_CH(reg) + 1;
2080 sck_way = TAD_SOCK(reg);
2082 if (ch_way == 3)
2083 idx = addr >> 6;
2084 else {
2085 idx = (addr >> (6 + sck_way + shiftup)) & 0x3;
2086 if (pvt->is_chan_hash)
2087 idx = haswell_chan_hash(idx, addr);
2089 idx = idx % ch_way;
2092 * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ???
2094 switch (idx) {
2095 case 0:
2096 base_ch = TAD_TGT0(reg);
2097 break;
2098 case 1:
2099 base_ch = TAD_TGT1(reg);
2100 break;
2101 case 2:
2102 base_ch = TAD_TGT2(reg);
2103 break;
2104 case 3:
2105 base_ch = TAD_TGT3(reg);
2106 break;
2107 default:
2108 sprintf(msg, "Can't discover the TAD target");
2109 return -EINVAL;
2111 *channel_mask = 1 << base_ch;
2113 pci_read_config_dword(pvt->pci_tad[base_ch], tad_ch_nilv_offset[n_tads], &tad_offset);
2115 if (pvt->mirror_mode == FULL_MIRRORING ||
2116 (pvt->mirror_mode == ADDR_RANGE_MIRRORING && n_tads == 0)) {
2117 *channel_mask |= 1 << ((base_ch + 2) % 4);
2118 switch(ch_way) {
2119 case 2:
2120 case 4:
2121 sck_xch = (1 << sck_way) * (ch_way >> 1);
2122 break;
2123 default:
2124 sprintf(msg, "Invalid mirror set. Can't decode addr");
2125 return -EINVAL;
2128 pvt->is_cur_addr_mirrored = true;
2129 } else {
2130 sck_xch = (1 << sck_way) * ch_way;
2131 pvt->is_cur_addr_mirrored = false;
2134 if (pvt->is_lockstep)
2135 *channel_mask |= 1 << ((base_ch + 1) % 4);
2137 offset = TAD_OFFSET(tad_offset);
2139 edac_dbg(0, "TAD#%d: address 0x%016Lx < 0x%016Lx, socket interleave %d, channel interleave %d (offset 0x%08Lx), index %d, base ch: %d, ch mask: 0x%02lx\n",
2140 n_tads,
2141 addr,
2142 limit,
2143 sck_way,
2144 ch_way,
2145 offset,
2146 idx,
2147 base_ch,
2148 *channel_mask);
2150 /* Calculate channel address */
2151 /* Remove the TAD offset */
2153 if (offset > addr) {
2154 sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!",
2155 offset, addr);
2156 return -EINVAL;
2159 ch_addr = addr - offset;
2160 ch_addr >>= (6 + shiftup);
2161 ch_addr /= sck_xch;
2162 ch_addr <<= (6 + shiftup);
2163 ch_addr |= addr & ((1 << (6 + shiftup)) - 1);
2166 * Step 3) Decode rank
2168 for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) {
2169 pci_read_config_dword(pvt->pci_tad[base_ch], rir_way_limit[n_rir], &reg);
2171 if (!IS_RIR_VALID(reg))
2172 continue;
2174 limit = pvt->info.rir_limit(reg);
2175 gb = div_u64_rem(limit >> 20, 1024, &mb);
2176 edac_dbg(0, "RIR#%d, limit: %u.%03u GB (0x%016Lx), way: %d\n",
2177 n_rir,
2178 gb, (mb*1000)/1024,
2179 limit,
2180 1 << RIR_WAY(reg));
2181 if (ch_addr <= limit)
2182 break;
2184 if (n_rir == MAX_RIR_RANGES) {
2185 sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx",
2186 ch_addr);
2187 return -EINVAL;
2189 rir_way = RIR_WAY(reg);
2191 if (pvt->is_close_pg)
2192 idx = (ch_addr >> 6);
2193 else
2194 idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */
2195 idx %= 1 << rir_way;
2197 pci_read_config_dword(pvt->pci_tad[base_ch], rir_offset[n_rir][idx], &reg);
2198 *rank = RIR_RNK_TGT(pvt->info.type, reg);
2200 edac_dbg(0, "RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n",
2201 n_rir,
2202 ch_addr,
2203 limit,
2204 rir_way,
2205 idx);
2207 return 0;
2210 /****************************************************************************
2211 Device initialization routines: put/get, init/exit
2212 ****************************************************************************/
2215 * sbridge_put_all_devices 'put' all the devices that we have
2216 * reserved via 'get'
2218 static void sbridge_put_devices(struct sbridge_dev *sbridge_dev)
2220 int i;
2222 edac_dbg(0, "\n");
2223 for (i = 0; i < sbridge_dev->n_devs; i++) {
2224 struct pci_dev *pdev = sbridge_dev->pdev[i];
2225 if (!pdev)
2226 continue;
2227 edac_dbg(0, "Removing dev %02x:%02x.%d\n",
2228 pdev->bus->number,
2229 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
2230 pci_dev_put(pdev);
2234 static void sbridge_put_all_devices(void)
2236 struct sbridge_dev *sbridge_dev, *tmp;
2238 list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) {
2239 sbridge_put_devices(sbridge_dev);
2240 free_sbridge_dev(sbridge_dev);
2244 static int sbridge_get_onedevice(struct pci_dev **prev,
2245 u8 *num_mc,
2246 const struct pci_id_table *table,
2247 const unsigned devno,
2248 const int multi_bus)
2250 struct sbridge_dev *sbridge_dev = NULL;
2251 const struct pci_id_descr *dev_descr = &table->descr[devno];
2252 struct pci_dev *pdev = NULL;
2253 int seg = 0;
2254 u8 bus = 0;
2255 int i = 0;
2257 sbridge_printk(KERN_DEBUG,
2258 "Seeking for: PCI ID %04x:%04x\n",
2259 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2261 pdev = pci_get_device(PCI_VENDOR_ID_INTEL,
2262 dev_descr->dev_id, *prev);
2264 if (!pdev) {
2265 if (*prev) {
2266 *prev = pdev;
2267 return 0;
2270 if (dev_descr->optional)
2271 return 0;
2273 /* if the HA wasn't found */
2274 if (devno == 0)
2275 return -ENODEV;
2277 sbridge_printk(KERN_INFO,
2278 "Device not found: %04x:%04x\n",
2279 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2281 /* End of list, leave */
2282 return -ENODEV;
2284 seg = pci_domain_nr(pdev->bus);
2285 bus = pdev->bus->number;
2287 next_imc:
2288 sbridge_dev = get_sbridge_dev(seg, bus, dev_descr->dom,
2289 multi_bus, sbridge_dev);
2290 if (!sbridge_dev) {
2291 /* If the HA1 wasn't found, don't create EDAC second memory controller */
2292 if (dev_descr->dom == IMC1 && devno != 1) {
2293 edac_dbg(0, "Skip IMC1: %04x:%04x (since HA1 was absent)\n",
2294 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2295 pci_dev_put(pdev);
2296 return 0;
2299 if (dev_descr->dom == SOCK)
2300 goto out_imc;
2302 sbridge_dev = alloc_sbridge_dev(seg, bus, dev_descr->dom, table);
2303 if (!sbridge_dev) {
2304 pci_dev_put(pdev);
2305 return -ENOMEM;
2307 (*num_mc)++;
2310 if (sbridge_dev->pdev[sbridge_dev->i_devs]) {
2311 sbridge_printk(KERN_ERR,
2312 "Duplicated device for %04x:%04x\n",
2313 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2314 pci_dev_put(pdev);
2315 return -ENODEV;
2318 sbridge_dev->pdev[sbridge_dev->i_devs++] = pdev;
2320 /* pdev belongs to more than one IMC, do extra gets */
2321 if (++i > 1)
2322 pci_dev_get(pdev);
2324 if (dev_descr->dom == SOCK && i < table->n_imcs_per_sock)
2325 goto next_imc;
2327 out_imc:
2328 /* Be sure that the device is enabled */
2329 if (unlikely(pci_enable_device(pdev) < 0)) {
2330 sbridge_printk(KERN_ERR,
2331 "Couldn't enable %04x:%04x\n",
2332 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2333 return -ENODEV;
2336 edac_dbg(0, "Detected %04x:%04x\n",
2337 PCI_VENDOR_ID_INTEL, dev_descr->dev_id);
2340 * As stated on drivers/pci/search.c, the reference count for
2341 * @from is always decremented if it is not %NULL. So, as we need
2342 * to get all devices up to null, we need to do a get for the device
2344 pci_dev_get(pdev);
2346 *prev = pdev;
2348 return 0;
2352 * sbridge_get_all_devices - Find and perform 'get' operation on the MCH's
2353 * devices we want to reference for this driver.
2354 * @num_mc: pointer to the memory controllers count, to be incremented in case
2355 * of success.
2356 * @table: model specific table
2358 * returns 0 in case of success or error code
2360 static int sbridge_get_all_devices(u8 *num_mc,
2361 const struct pci_id_table *table)
2363 int i, rc;
2364 struct pci_dev *pdev = NULL;
2365 int allow_dups = 0;
2366 int multi_bus = 0;
2368 if (table->type == KNIGHTS_LANDING)
2369 allow_dups = multi_bus = 1;
2370 while (table && table->descr) {
2371 for (i = 0; i < table->n_devs_per_sock; i++) {
2372 if (!allow_dups || i == 0 ||
2373 table->descr[i].dev_id !=
2374 table->descr[i-1].dev_id) {
2375 pdev = NULL;
2377 do {
2378 rc = sbridge_get_onedevice(&pdev, num_mc,
2379 table, i, multi_bus);
2380 if (rc < 0) {
2381 if (i == 0) {
2382 i = table->n_devs_per_sock;
2383 break;
2385 sbridge_put_all_devices();
2386 return -ENODEV;
2388 } while (pdev && !allow_dups);
2390 table++;
2393 return 0;
2397 * Device IDs for {SBRIDGE,IBRIDGE,HASWELL,BROADWELL}_IMC_HA0_TAD0 are in
2398 * the format: XXXa. So we can convert from a device to the corresponding
2399 * channel like this
2401 #define TAD_DEV_TO_CHAN(dev) (((dev) & 0xf) - 0xa)
2403 static int sbridge_mci_bind_devs(struct mem_ctl_info *mci,
2404 struct sbridge_dev *sbridge_dev)
2406 struct sbridge_pvt *pvt = mci->pvt_info;
2407 struct pci_dev *pdev;
2408 u8 saw_chan_mask = 0;
2409 int i;
2411 for (i = 0; i < sbridge_dev->n_devs; i++) {
2412 pdev = sbridge_dev->pdev[i];
2413 if (!pdev)
2414 continue;
2416 switch (pdev->device) {
2417 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0:
2418 pvt->pci_sad0 = pdev;
2419 break;
2420 case PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1:
2421 pvt->pci_sad1 = pdev;
2422 break;
2423 case PCI_DEVICE_ID_INTEL_SBRIDGE_BR:
2424 pvt->pci_br0 = pdev;
2425 break;
2426 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0:
2427 pvt->pci_ha = pdev;
2428 break;
2429 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA:
2430 pvt->pci_ta = pdev;
2431 break;
2432 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS:
2433 pvt->pci_ras = pdev;
2434 break;
2435 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0:
2436 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1:
2437 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2:
2438 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3:
2440 int id = TAD_DEV_TO_CHAN(pdev->device);
2441 pvt->pci_tad[id] = pdev;
2442 saw_chan_mask |= 1 << id;
2444 break;
2445 case PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO:
2446 pvt->pci_ddrio = pdev;
2447 break;
2448 default:
2449 goto error;
2452 edac_dbg(0, "Associated PCI %02x:%02x, bus %d with dev = %p\n",
2453 pdev->vendor, pdev->device,
2454 sbridge_dev->bus,
2455 pdev);
2458 /* Check if everything were registered */
2459 if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha ||
2460 !pvt->pci_ras || !pvt->pci_ta)
2461 goto enodev;
2463 if (saw_chan_mask != 0x0f)
2464 goto enodev;
2465 return 0;
2467 enodev:
2468 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2469 return -ENODEV;
2471 error:
2472 sbridge_printk(KERN_ERR, "Unexpected device %02x:%02x\n",
2473 PCI_VENDOR_ID_INTEL, pdev->device);
2474 return -EINVAL;
2477 static int ibridge_mci_bind_devs(struct mem_ctl_info *mci,
2478 struct sbridge_dev *sbridge_dev)
2480 struct sbridge_pvt *pvt = mci->pvt_info;
2481 struct pci_dev *pdev;
2482 u8 saw_chan_mask = 0;
2483 int i;
2485 for (i = 0; i < sbridge_dev->n_devs; i++) {
2486 pdev = sbridge_dev->pdev[i];
2487 if (!pdev)
2488 continue;
2490 switch (pdev->device) {
2491 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0:
2492 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1:
2493 pvt->pci_ha = pdev;
2494 break;
2495 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TA:
2496 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TA:
2497 pvt->pci_ta = pdev;
2498 break;
2499 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_RAS:
2500 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_RAS:
2501 pvt->pci_ras = pdev;
2502 break;
2503 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD0:
2504 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD1:
2505 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD2:
2506 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA0_TAD3:
2507 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD0:
2508 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD1:
2509 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD2:
2510 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_HA1_TAD3:
2512 int id = TAD_DEV_TO_CHAN(pdev->device);
2513 pvt->pci_tad[id] = pdev;
2514 saw_chan_mask |= 1 << id;
2516 break;
2517 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_2HA_DDRIO0:
2518 pvt->pci_ddrio = pdev;
2519 break;
2520 case PCI_DEVICE_ID_INTEL_IBRIDGE_IMC_1HA_DDRIO0:
2521 pvt->pci_ddrio = pdev;
2522 break;
2523 case PCI_DEVICE_ID_INTEL_IBRIDGE_SAD:
2524 pvt->pci_sad0 = pdev;
2525 break;
2526 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR0:
2527 pvt->pci_br0 = pdev;
2528 break;
2529 case PCI_DEVICE_ID_INTEL_IBRIDGE_BR1:
2530 pvt->pci_br1 = pdev;
2531 break;
2532 default:
2533 goto error;
2536 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2537 sbridge_dev->bus,
2538 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2539 pdev);
2542 /* Check if everything were registered */
2543 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_br0 ||
2544 !pvt->pci_br1 || !pvt->pci_ras || !pvt->pci_ta)
2545 goto enodev;
2547 if (saw_chan_mask != 0x0f && /* -EN/-EX */
2548 saw_chan_mask != 0x03) /* -EP */
2549 goto enodev;
2550 return 0;
2552 enodev:
2553 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2554 return -ENODEV;
2556 error:
2557 sbridge_printk(KERN_ERR,
2558 "Unexpected device %02x:%02x\n", PCI_VENDOR_ID_INTEL,
2559 pdev->device);
2560 return -EINVAL;
2563 static int haswell_mci_bind_devs(struct mem_ctl_info *mci,
2564 struct sbridge_dev *sbridge_dev)
2566 struct sbridge_pvt *pvt = mci->pvt_info;
2567 struct pci_dev *pdev;
2568 u8 saw_chan_mask = 0;
2569 int i;
2571 /* there's only one device per system; not tied to any bus */
2572 if (pvt->info.pci_vtd == NULL)
2573 /* result will be checked later */
2574 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2575 PCI_DEVICE_ID_INTEL_HASWELL_IMC_VTD_MISC,
2576 NULL);
2578 for (i = 0; i < sbridge_dev->n_devs; i++) {
2579 pdev = sbridge_dev->pdev[i];
2580 if (!pdev)
2581 continue;
2583 switch (pdev->device) {
2584 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD0:
2585 pvt->pci_sad0 = pdev;
2586 break;
2587 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_CBO_SAD1:
2588 pvt->pci_sad1 = pdev;
2589 break;
2590 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0:
2591 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1:
2592 pvt->pci_ha = pdev;
2593 break;
2594 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TA:
2595 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TA:
2596 pvt->pci_ta = pdev;
2597 break;
2598 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TM:
2599 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TM:
2600 pvt->pci_ras = pdev;
2601 break;
2602 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD0:
2603 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD1:
2604 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD2:
2605 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA0_TAD3:
2606 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD0:
2607 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD1:
2608 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD2:
2609 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_HA1_TAD3:
2611 int id = TAD_DEV_TO_CHAN(pdev->device);
2612 pvt->pci_tad[id] = pdev;
2613 saw_chan_mask |= 1 << id;
2615 break;
2616 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO0:
2617 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO1:
2618 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO2:
2619 case PCI_DEVICE_ID_INTEL_HASWELL_IMC_DDRIO3:
2620 if (!pvt->pci_ddrio)
2621 pvt->pci_ddrio = pdev;
2622 break;
2623 default:
2624 break;
2627 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2628 sbridge_dev->bus,
2629 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2630 pdev);
2633 /* Check if everything were registered */
2634 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_sad1 ||
2635 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2636 goto enodev;
2638 if (saw_chan_mask != 0x0f && /* -EN/-EX */
2639 saw_chan_mask != 0x03) /* -EP */
2640 goto enodev;
2641 return 0;
2643 enodev:
2644 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2645 return -ENODEV;
2648 static int broadwell_mci_bind_devs(struct mem_ctl_info *mci,
2649 struct sbridge_dev *sbridge_dev)
2651 struct sbridge_pvt *pvt = mci->pvt_info;
2652 struct pci_dev *pdev;
2653 u8 saw_chan_mask = 0;
2654 int i;
2656 /* there's only one device per system; not tied to any bus */
2657 if (pvt->info.pci_vtd == NULL)
2658 /* result will be checked later */
2659 pvt->info.pci_vtd = pci_get_device(PCI_VENDOR_ID_INTEL,
2660 PCI_DEVICE_ID_INTEL_BROADWELL_IMC_VTD_MISC,
2661 NULL);
2663 for (i = 0; i < sbridge_dev->n_devs; i++) {
2664 pdev = sbridge_dev->pdev[i];
2665 if (!pdev)
2666 continue;
2668 switch (pdev->device) {
2669 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD0:
2670 pvt->pci_sad0 = pdev;
2671 break;
2672 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_CBO_SAD1:
2673 pvt->pci_sad1 = pdev;
2674 break;
2675 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0:
2676 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1:
2677 pvt->pci_ha = pdev;
2678 break;
2679 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TA:
2680 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TA:
2681 pvt->pci_ta = pdev;
2682 break;
2683 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TM:
2684 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TM:
2685 pvt->pci_ras = pdev;
2686 break;
2687 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD0:
2688 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD1:
2689 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD2:
2690 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA0_TAD3:
2691 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD0:
2692 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD1:
2693 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD2:
2694 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_HA1_TAD3:
2696 int id = TAD_DEV_TO_CHAN(pdev->device);
2697 pvt->pci_tad[id] = pdev;
2698 saw_chan_mask |= 1 << id;
2700 break;
2701 case PCI_DEVICE_ID_INTEL_BROADWELL_IMC_DDRIO0:
2702 pvt->pci_ddrio = pdev;
2703 break;
2704 default:
2705 break;
2708 edac_dbg(0, "Associated PCI %02x.%02d.%d with dev = %p\n",
2709 sbridge_dev->bus,
2710 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2711 pdev);
2714 /* Check if everything were registered */
2715 if (!pvt->pci_sad0 || !pvt->pci_ha || !pvt->pci_sad1 ||
2716 !pvt->pci_ras || !pvt->pci_ta || !pvt->info.pci_vtd)
2717 goto enodev;
2719 if (saw_chan_mask != 0x0f && /* -EN/-EX */
2720 saw_chan_mask != 0x03) /* -EP */
2721 goto enodev;
2722 return 0;
2724 enodev:
2725 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2726 return -ENODEV;
2729 static int knl_mci_bind_devs(struct mem_ctl_info *mci,
2730 struct sbridge_dev *sbridge_dev)
2732 struct sbridge_pvt *pvt = mci->pvt_info;
2733 struct pci_dev *pdev;
2734 int dev, func;
2736 int i;
2737 int devidx;
2739 for (i = 0; i < sbridge_dev->n_devs; i++) {
2740 pdev = sbridge_dev->pdev[i];
2741 if (!pdev)
2742 continue;
2744 /* Extract PCI device and function. */
2745 dev = (pdev->devfn >> 3) & 0x1f;
2746 func = pdev->devfn & 0x7;
2748 switch (pdev->device) {
2749 case PCI_DEVICE_ID_INTEL_KNL_IMC_MC:
2750 if (dev == 8)
2751 pvt->knl.pci_mc0 = pdev;
2752 else if (dev == 9)
2753 pvt->knl.pci_mc1 = pdev;
2754 else {
2755 sbridge_printk(KERN_ERR,
2756 "Memory controller in unexpected place! (dev %d, fn %d)\n",
2757 dev, func);
2758 continue;
2760 break;
2762 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD0:
2763 pvt->pci_sad0 = pdev;
2764 break;
2766 case PCI_DEVICE_ID_INTEL_KNL_IMC_SAD1:
2767 pvt->pci_sad1 = pdev;
2768 break;
2770 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHA:
2771 /* There are one of these per tile, and range from
2772 * 1.14.0 to 1.18.5.
2774 devidx = ((dev-14)*8)+func;
2776 if (devidx < 0 || devidx >= KNL_MAX_CHAS) {
2777 sbridge_printk(KERN_ERR,
2778 "Caching and Home Agent in unexpected place! (dev %d, fn %d)\n",
2779 dev, func);
2780 continue;
2783 WARN_ON(pvt->knl.pci_cha[devidx] != NULL);
2785 pvt->knl.pci_cha[devidx] = pdev;
2786 break;
2788 case PCI_DEVICE_ID_INTEL_KNL_IMC_CHAN:
2789 devidx = -1;
2792 * MC0 channels 0-2 are device 9 function 2-4,
2793 * MC1 channels 3-5 are device 8 function 2-4.
2796 if (dev == 9)
2797 devidx = func-2;
2798 else if (dev == 8)
2799 devidx = 3 + (func-2);
2801 if (devidx < 0 || devidx >= KNL_MAX_CHANNELS) {
2802 sbridge_printk(KERN_ERR,
2803 "DRAM Channel Registers in unexpected place! (dev %d, fn %d)\n",
2804 dev, func);
2805 continue;
2808 WARN_ON(pvt->knl.pci_channel[devidx] != NULL);
2809 pvt->knl.pci_channel[devidx] = pdev;
2810 break;
2812 case PCI_DEVICE_ID_INTEL_KNL_IMC_TOLHM:
2813 pvt->knl.pci_mc_info = pdev;
2814 break;
2816 case PCI_DEVICE_ID_INTEL_KNL_IMC_TA:
2817 pvt->pci_ta = pdev;
2818 break;
2820 default:
2821 sbridge_printk(KERN_ERR, "Unexpected device %d\n",
2822 pdev->device);
2823 break;
2827 if (!pvt->knl.pci_mc0 || !pvt->knl.pci_mc1 ||
2828 !pvt->pci_sad0 || !pvt->pci_sad1 ||
2829 !pvt->pci_ta) {
2830 goto enodev;
2833 for (i = 0; i < KNL_MAX_CHANNELS; i++) {
2834 if (!pvt->knl.pci_channel[i]) {
2835 sbridge_printk(KERN_ERR, "Missing channel %d\n", i);
2836 goto enodev;
2840 for (i = 0; i < KNL_MAX_CHAS; i++) {
2841 if (!pvt->knl.pci_cha[i]) {
2842 sbridge_printk(KERN_ERR, "Missing CHA %d\n", i);
2843 goto enodev;
2847 return 0;
2849 enodev:
2850 sbridge_printk(KERN_ERR, "Some needed devices are missing\n");
2851 return -ENODEV;
2854 /****************************************************************************
2855 Error check routines
2856 ****************************************************************************/
2859 * While Sandy Bridge has error count registers, SMI BIOS read values from
2860 * and resets the counters. So, they are not reliable for the OS to read
2861 * from them. So, we have no option but to just trust on whatever MCE is
2862 * telling us about the errors.
2864 static void sbridge_mce_output_error(struct mem_ctl_info *mci,
2865 const struct mce *m)
2867 struct mem_ctl_info *new_mci;
2868 struct sbridge_pvt *pvt = mci->pvt_info;
2869 enum hw_event_mc_err_type tp_event;
2870 char *type, *optype, msg[256];
2871 bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0);
2872 bool overflow = GET_BITFIELD(m->status, 62, 62);
2873 bool uncorrected_error = GET_BITFIELD(m->status, 61, 61);
2874 bool recoverable;
2875 u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52);
2876 u32 mscod = GET_BITFIELD(m->status, 16, 31);
2877 u32 errcode = GET_BITFIELD(m->status, 0, 15);
2878 u32 channel = GET_BITFIELD(m->status, 0, 3);
2879 u32 optypenum = GET_BITFIELD(m->status, 4, 6);
2880 long channel_mask, first_channel;
2881 u8 rank, socket, ha;
2882 int rc, dimm;
2883 char *area_type = NULL;
2885 if (pvt->info.type != SANDY_BRIDGE)
2886 recoverable = true;
2887 else
2888 recoverable = GET_BITFIELD(m->status, 56, 56);
2890 if (uncorrected_error) {
2891 if (ripv) {
2892 type = "FATAL";
2893 tp_event = HW_EVENT_ERR_FATAL;
2894 } else {
2895 type = "NON_FATAL";
2896 tp_event = HW_EVENT_ERR_UNCORRECTED;
2898 } else {
2899 type = "CORRECTED";
2900 tp_event = HW_EVENT_ERR_CORRECTED;
2904 * According with Table 15-9 of the Intel Architecture spec vol 3A,
2905 * memory errors should fit in this mask:
2906 * 000f 0000 1mmm cccc (binary)
2907 * where:
2908 * f = Correction Report Filtering Bit. If 1, subsequent errors
2909 * won't be shown
2910 * mmm = error type
2911 * cccc = channel
2912 * If the mask doesn't match, report an error to the parsing logic
2914 if (! ((errcode & 0xef80) == 0x80)) {
2915 optype = "Can't parse: it is not a mem";
2916 } else {
2917 switch (optypenum) {
2918 case 0:
2919 optype = "generic undef request error";
2920 break;
2921 case 1:
2922 optype = "memory read error";
2923 break;
2924 case 2:
2925 optype = "memory write error";
2926 break;
2927 case 3:
2928 optype = "addr/cmd error";
2929 break;
2930 case 4:
2931 optype = "memory scrubbing error";
2932 break;
2933 default:
2934 optype = "reserved";
2935 break;
2939 /* Only decode errors with an valid address (ADDRV) */
2940 if (!GET_BITFIELD(m->status, 58, 58))
2941 return;
2943 if (pvt->info.type == KNIGHTS_LANDING) {
2944 if (channel == 14) {
2945 edac_dbg(0, "%s%s err_code:%04x:%04x EDRAM bank %d\n",
2946 overflow ? " OVERFLOW" : "",
2947 (uncorrected_error && recoverable)
2948 ? " recoverable" : "",
2949 mscod, errcode,
2950 m->bank);
2951 } else {
2952 char A = *("A");
2955 * Reported channel is in range 0-2, so we can't map it
2956 * back to mc. To figure out mc we check machine check
2957 * bank register that reported this error.
2958 * bank15 means mc0 and bank16 means mc1.
2960 channel = knl_channel_remap(m->bank == 16, channel);
2961 channel_mask = 1 << channel;
2963 snprintf(msg, sizeof(msg),
2964 "%s%s err_code:%04x:%04x channel:%d (DIMM_%c)",
2965 overflow ? " OVERFLOW" : "",
2966 (uncorrected_error && recoverable)
2967 ? " recoverable" : " ",
2968 mscod, errcode, channel, A + channel);
2969 edac_mc_handle_error(tp_event, mci, core_err_cnt,
2970 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
2971 channel, 0, -1,
2972 optype, msg);
2974 return;
2975 } else {
2976 rc = get_memory_error_data(mci, m->addr, &socket, &ha,
2977 &channel_mask, &rank, &area_type, msg);
2980 if (rc < 0)
2981 goto err_parsing;
2982 new_mci = get_mci_for_node_id(socket, ha);
2983 if (!new_mci) {
2984 strcpy(msg, "Error: socket got corrupted!");
2985 goto err_parsing;
2987 mci = new_mci;
2988 pvt = mci->pvt_info;
2990 first_channel = find_first_bit(&channel_mask, NUM_CHANNELS);
2992 if (rank < 4)
2993 dimm = 0;
2994 else if (rank < 8)
2995 dimm = 1;
2996 else
2997 dimm = 2;
3001 * FIXME: On some memory configurations (mirror, lockstep), the
3002 * Memory Controller can't point the error to a single DIMM. The
3003 * EDAC core should be handling the channel mask, in order to point
3004 * to the group of dimm's where the error may be happening.
3006 if (!pvt->is_lockstep && !pvt->is_cur_addr_mirrored && !pvt->is_close_pg)
3007 channel = first_channel;
3009 snprintf(msg, sizeof(msg),
3010 "%s%s area:%s err_code:%04x:%04x socket:%d ha:%d channel_mask:%ld rank:%d",
3011 overflow ? " OVERFLOW" : "",
3012 (uncorrected_error && recoverable) ? " recoverable" : "",
3013 area_type,
3014 mscod, errcode,
3015 socket, ha,
3016 channel_mask,
3017 rank);
3019 edac_dbg(0, "%s\n", msg);
3021 /* FIXME: need support for channel mask */
3023 if (channel == CHANNEL_UNSPECIFIED)
3024 channel = -1;
3026 /* Call the helper to output message */
3027 edac_mc_handle_error(tp_event, mci, core_err_cnt,
3028 m->addr >> PAGE_SHIFT, m->addr & ~PAGE_MASK, 0,
3029 channel, dimm, -1,
3030 optype, msg);
3031 return;
3032 err_parsing:
3033 edac_mc_handle_error(tp_event, mci, core_err_cnt, 0, 0, 0,
3034 -1, -1, -1,
3035 msg, "");
3040 * Check that logging is enabled and that this is the right type
3041 * of error for us to handle.
3043 static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val,
3044 void *data)
3046 struct mce *mce = (struct mce *)data;
3047 struct mem_ctl_info *mci;
3048 struct sbridge_pvt *pvt;
3049 char *type;
3051 if (edac_get_report_status() == EDAC_REPORTING_DISABLED)
3052 return NOTIFY_DONE;
3054 mci = get_mci_for_node_id(mce->socketid, IMC0);
3055 if (!mci)
3056 return NOTIFY_DONE;
3057 pvt = mci->pvt_info;
3060 * Just let mcelog handle it if the error is
3061 * outside the memory controller. A memory error
3062 * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0.
3063 * bit 12 has an special meaning.
3065 if ((mce->status & 0xefff) >> 7 != 1)
3066 return NOTIFY_DONE;
3068 if (mce->mcgstatus & MCG_STATUS_MCIP)
3069 type = "Exception";
3070 else
3071 type = "Event";
3073 sbridge_mc_printk(mci, KERN_DEBUG, "HANDLING MCE MEMORY ERROR\n");
3075 sbridge_mc_printk(mci, KERN_DEBUG, "CPU %d: Machine Check %s: %Lx "
3076 "Bank %d: %016Lx\n", mce->extcpu, type,
3077 mce->mcgstatus, mce->bank, mce->status);
3078 sbridge_mc_printk(mci, KERN_DEBUG, "TSC %llx ", mce->tsc);
3079 sbridge_mc_printk(mci, KERN_DEBUG, "ADDR %llx ", mce->addr);
3080 sbridge_mc_printk(mci, KERN_DEBUG, "MISC %llx ", mce->misc);
3082 sbridge_mc_printk(mci, KERN_DEBUG, "PROCESSOR %u:%x TIME %llu SOCKET "
3083 "%u APIC %x\n", mce->cpuvendor, mce->cpuid,
3084 mce->time, mce->socketid, mce->apicid);
3086 sbridge_mce_output_error(mci, mce);
3088 /* Advice mcelog that the error were handled */
3089 return NOTIFY_STOP;
3092 static struct notifier_block sbridge_mce_dec = {
3093 .notifier_call = sbridge_mce_check_error,
3094 .priority = MCE_PRIO_EDAC,
3097 /****************************************************************************
3098 EDAC register/unregister logic
3099 ****************************************************************************/
3101 static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev)
3103 struct mem_ctl_info *mci = sbridge_dev->mci;
3104 struct sbridge_pvt *pvt;
3106 if (unlikely(!mci || !mci->pvt_info)) {
3107 edac_dbg(0, "MC: dev = %p\n", &sbridge_dev->pdev[0]->dev);
3109 sbridge_printk(KERN_ERR, "Couldn't find mci handler\n");
3110 return;
3113 pvt = mci->pvt_info;
3115 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3116 mci, &sbridge_dev->pdev[0]->dev);
3118 /* Remove MC sysfs nodes */
3119 edac_mc_del_mc(mci->pdev);
3121 edac_dbg(1, "%s: free mci struct\n", mci->ctl_name);
3122 kfree(mci->ctl_name);
3123 edac_mc_free(mci);
3124 sbridge_dev->mci = NULL;
3127 static int sbridge_register_mci(struct sbridge_dev *sbridge_dev, enum type type)
3129 struct mem_ctl_info *mci;
3130 struct edac_mc_layer layers[2];
3131 struct sbridge_pvt *pvt;
3132 struct pci_dev *pdev = sbridge_dev->pdev[0];
3133 int rc;
3135 /* allocate a new MC control structure */
3136 layers[0].type = EDAC_MC_LAYER_CHANNEL;
3137 layers[0].size = type == KNIGHTS_LANDING ?
3138 KNL_MAX_CHANNELS : NUM_CHANNELS;
3139 layers[0].is_virt_csrow = false;
3140 layers[1].type = EDAC_MC_LAYER_SLOT;
3141 layers[1].size = type == KNIGHTS_LANDING ? 1 : MAX_DIMMS;
3142 layers[1].is_virt_csrow = true;
3143 mci = edac_mc_alloc(sbridge_dev->mc, ARRAY_SIZE(layers), layers,
3144 sizeof(*pvt));
3146 if (unlikely(!mci))
3147 return -ENOMEM;
3149 edac_dbg(0, "MC: mci = %p, dev = %p\n",
3150 mci, &pdev->dev);
3152 pvt = mci->pvt_info;
3153 memset(pvt, 0, sizeof(*pvt));
3155 /* Associate sbridge_dev and mci for future usage */
3156 pvt->sbridge_dev = sbridge_dev;
3157 sbridge_dev->mci = mci;
3159 mci->mtype_cap = type == KNIGHTS_LANDING ?
3160 MEM_FLAG_DDR4 : MEM_FLAG_DDR3;
3161 mci->edac_ctl_cap = EDAC_FLAG_NONE;
3162 mci->edac_cap = EDAC_FLAG_NONE;
3163 mci->mod_name = EDAC_MOD_STR;
3164 mci->dev_name = pci_name(pdev);
3165 mci->ctl_page_to_phys = NULL;
3167 pvt->info.type = type;
3168 switch (type) {
3169 case IVY_BRIDGE:
3170 pvt->info.rankcfgr = IB_RANK_CFG_A;
3171 pvt->info.get_tolm = ibridge_get_tolm;
3172 pvt->info.get_tohm = ibridge_get_tohm;
3173 pvt->info.dram_rule = ibridge_dram_rule;
3174 pvt->info.get_memory_type = get_memory_type;
3175 pvt->info.get_node_id = get_node_id;
3176 pvt->info.rir_limit = rir_limit;
3177 pvt->info.sad_limit = sad_limit;
3178 pvt->info.interleave_mode = interleave_mode;
3179 pvt->info.dram_attr = dram_attr;
3180 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3181 pvt->info.interleave_list = ibridge_interleave_list;
3182 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3183 pvt->info.get_width = ibridge_get_width;
3185 /* Store pci devices at mci for faster access */
3186 rc = ibridge_mci_bind_devs(mci, sbridge_dev);
3187 if (unlikely(rc < 0))
3188 goto fail0;
3189 get_source_id(mci);
3190 mci->ctl_name = kasprintf(GFP_KERNEL, "Ivy Bridge SrcID#%d_Ha#%d",
3191 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3192 break;
3193 case SANDY_BRIDGE:
3194 pvt->info.rankcfgr = SB_RANK_CFG_A;
3195 pvt->info.get_tolm = sbridge_get_tolm;
3196 pvt->info.get_tohm = sbridge_get_tohm;
3197 pvt->info.dram_rule = sbridge_dram_rule;
3198 pvt->info.get_memory_type = get_memory_type;
3199 pvt->info.get_node_id = get_node_id;
3200 pvt->info.rir_limit = rir_limit;
3201 pvt->info.sad_limit = sad_limit;
3202 pvt->info.interleave_mode = interleave_mode;
3203 pvt->info.dram_attr = dram_attr;
3204 pvt->info.max_sad = ARRAY_SIZE(sbridge_dram_rule);
3205 pvt->info.interleave_list = sbridge_interleave_list;
3206 pvt->info.interleave_pkg = sbridge_interleave_pkg;
3207 pvt->info.get_width = sbridge_get_width;
3209 /* Store pci devices at mci for faster access */
3210 rc = sbridge_mci_bind_devs(mci, sbridge_dev);
3211 if (unlikely(rc < 0))
3212 goto fail0;
3213 get_source_id(mci);
3214 mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge SrcID#%d_Ha#%d",
3215 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3216 break;
3217 case HASWELL:
3218 /* rankcfgr isn't used */
3219 pvt->info.get_tolm = haswell_get_tolm;
3220 pvt->info.get_tohm = haswell_get_tohm;
3221 pvt->info.dram_rule = ibridge_dram_rule;
3222 pvt->info.get_memory_type = haswell_get_memory_type;
3223 pvt->info.get_node_id = haswell_get_node_id;
3224 pvt->info.rir_limit = haswell_rir_limit;
3225 pvt->info.sad_limit = sad_limit;
3226 pvt->info.interleave_mode = interleave_mode;
3227 pvt->info.dram_attr = dram_attr;
3228 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3229 pvt->info.interleave_list = ibridge_interleave_list;
3230 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3231 pvt->info.get_width = ibridge_get_width;
3233 /* Store pci devices at mci for faster access */
3234 rc = haswell_mci_bind_devs(mci, sbridge_dev);
3235 if (unlikely(rc < 0))
3236 goto fail0;
3237 get_source_id(mci);
3238 mci->ctl_name = kasprintf(GFP_KERNEL, "Haswell SrcID#%d_Ha#%d",
3239 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3240 break;
3241 case BROADWELL:
3242 /* rankcfgr isn't used */
3243 pvt->info.get_tolm = haswell_get_tolm;
3244 pvt->info.get_tohm = haswell_get_tohm;
3245 pvt->info.dram_rule = ibridge_dram_rule;
3246 pvt->info.get_memory_type = haswell_get_memory_type;
3247 pvt->info.get_node_id = haswell_get_node_id;
3248 pvt->info.rir_limit = haswell_rir_limit;
3249 pvt->info.sad_limit = sad_limit;
3250 pvt->info.interleave_mode = interleave_mode;
3251 pvt->info.dram_attr = dram_attr;
3252 pvt->info.max_sad = ARRAY_SIZE(ibridge_dram_rule);
3253 pvt->info.interleave_list = ibridge_interleave_list;
3254 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3255 pvt->info.get_width = broadwell_get_width;
3257 /* Store pci devices at mci for faster access */
3258 rc = broadwell_mci_bind_devs(mci, sbridge_dev);
3259 if (unlikely(rc < 0))
3260 goto fail0;
3261 get_source_id(mci);
3262 mci->ctl_name = kasprintf(GFP_KERNEL, "Broadwell SrcID#%d_Ha#%d",
3263 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3264 break;
3265 case KNIGHTS_LANDING:
3266 /* pvt->info.rankcfgr == ??? */
3267 pvt->info.get_tolm = knl_get_tolm;
3268 pvt->info.get_tohm = knl_get_tohm;
3269 pvt->info.dram_rule = knl_dram_rule;
3270 pvt->info.get_memory_type = knl_get_memory_type;
3271 pvt->info.get_node_id = knl_get_node_id;
3272 pvt->info.rir_limit = NULL;
3273 pvt->info.sad_limit = knl_sad_limit;
3274 pvt->info.interleave_mode = knl_interleave_mode;
3275 pvt->info.dram_attr = dram_attr_knl;
3276 pvt->info.max_sad = ARRAY_SIZE(knl_dram_rule);
3277 pvt->info.interleave_list = knl_interleave_list;
3278 pvt->info.interleave_pkg = ibridge_interleave_pkg;
3279 pvt->info.get_width = knl_get_width;
3281 rc = knl_mci_bind_devs(mci, sbridge_dev);
3282 if (unlikely(rc < 0))
3283 goto fail0;
3284 get_source_id(mci);
3285 mci->ctl_name = kasprintf(GFP_KERNEL, "Knights Landing SrcID#%d_Ha#%d",
3286 pvt->sbridge_dev->source_id, pvt->sbridge_dev->dom);
3287 break;
3290 if (!mci->ctl_name) {
3291 rc = -ENOMEM;
3292 goto fail0;
3295 /* Get dimm basic config and the memory layout */
3296 rc = get_dimm_config(mci);
3297 if (rc < 0) {
3298 edac_dbg(0, "MC: failed to get_dimm_config()\n");
3299 goto fail;
3301 get_memory_layout(mci);
3303 /* record ptr to the generic device */
3304 mci->pdev = &pdev->dev;
3306 /* add this new MC control structure to EDAC's list of MCs */
3307 if (unlikely(edac_mc_add_mc(mci))) {
3308 edac_dbg(0, "MC: failed edac_mc_add_mc()\n");
3309 rc = -EINVAL;
3310 goto fail;
3313 return 0;
3315 fail:
3316 kfree(mci->ctl_name);
3317 fail0:
3318 edac_mc_free(mci);
3319 sbridge_dev->mci = NULL;
3320 return rc;
3323 #define ICPU(model, table) \
3324 { X86_VENDOR_INTEL, 6, model, 0, (unsigned long)&table }
3326 static const struct x86_cpu_id sbridge_cpuids[] = {
3327 ICPU(INTEL_FAM6_SANDYBRIDGE_X, pci_dev_descr_sbridge_table),
3328 ICPU(INTEL_FAM6_IVYBRIDGE_X, pci_dev_descr_ibridge_table),
3329 ICPU(INTEL_FAM6_HASWELL_X, pci_dev_descr_haswell_table),
3330 ICPU(INTEL_FAM6_BROADWELL_X, pci_dev_descr_broadwell_table),
3331 ICPU(INTEL_FAM6_BROADWELL_XEON_D, pci_dev_descr_broadwell_table),
3332 ICPU(INTEL_FAM6_XEON_PHI_KNL, pci_dev_descr_knl_table),
3333 ICPU(INTEL_FAM6_XEON_PHI_KNM, pci_dev_descr_knl_table),
3336 MODULE_DEVICE_TABLE(x86cpu, sbridge_cpuids);
3339 * sbridge_probe Get all devices and register memory controllers
3340 * present.
3341 * return:
3342 * 0 for FOUND a device
3343 * < 0 for error code
3346 static int sbridge_probe(const struct x86_cpu_id *id)
3348 int rc = -ENODEV;
3349 u8 mc, num_mc = 0;
3350 struct sbridge_dev *sbridge_dev;
3351 struct pci_id_table *ptable = (struct pci_id_table *)id->driver_data;
3353 /* get the pci devices we want to reserve for our use */
3354 rc = sbridge_get_all_devices(&num_mc, ptable);
3356 if (unlikely(rc < 0)) {
3357 edac_dbg(0, "couldn't get all devices\n");
3358 goto fail0;
3361 mc = 0;
3363 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) {
3364 edac_dbg(0, "Registering MC#%d (%d of %d)\n",
3365 mc, mc + 1, num_mc);
3367 sbridge_dev->mc = mc++;
3368 rc = sbridge_register_mci(sbridge_dev, ptable->type);
3369 if (unlikely(rc < 0))
3370 goto fail1;
3373 sbridge_printk(KERN_INFO, "%s\n", SBRIDGE_REVISION);
3375 return 0;
3377 fail1:
3378 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3379 sbridge_unregister_mci(sbridge_dev);
3381 sbridge_put_all_devices();
3382 fail0:
3383 return rc;
3387 * sbridge_remove cleanup
3390 static void sbridge_remove(void)
3392 struct sbridge_dev *sbridge_dev;
3394 edac_dbg(0, "\n");
3396 list_for_each_entry(sbridge_dev, &sbridge_edac_list, list)
3397 sbridge_unregister_mci(sbridge_dev);
3399 /* Release PCI resources */
3400 sbridge_put_all_devices();
3404 * sbridge_init Module entry function
3405 * Try to initialize this module for its devices
3407 static int __init sbridge_init(void)
3409 const struct x86_cpu_id *id;
3410 const char *owner;
3411 int rc;
3413 edac_dbg(2, "\n");
3415 owner = edac_get_owner();
3416 if (owner && strncmp(owner, EDAC_MOD_STR, sizeof(EDAC_MOD_STR)))
3417 return -EBUSY;
3419 id = x86_match_cpu(sbridge_cpuids);
3420 if (!id)
3421 return -ENODEV;
3423 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
3424 opstate_init();
3426 rc = sbridge_probe(id);
3428 if (rc >= 0) {
3429 mce_register_decode_chain(&sbridge_mce_dec);
3430 if (edac_get_report_status() == EDAC_REPORTING_DISABLED)
3431 sbridge_printk(KERN_WARNING, "Loading driver, error reporting disabled.\n");
3432 return 0;
3435 sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n",
3436 rc);
3438 return rc;
3442 * sbridge_exit() Module exit function
3443 * Unregister the driver
3445 static void __exit sbridge_exit(void)
3447 edac_dbg(2, "\n");
3448 sbridge_remove();
3449 mce_unregister_decode_chain(&sbridge_mce_dec);
3452 module_init(sbridge_init);
3453 module_exit(sbridge_exit);
3455 module_param(edac_op_state, int, 0444);
3456 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");
3458 MODULE_LICENSE("GPL");
3459 MODULE_AUTHOR("Mauro Carvalho Chehab");
3460 MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)");
3461 MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge and Ivy Bridge memory controllers - "
3462 SBRIDGE_REVISION);